SALIENT SOCIO-ECONOMIC AND DEMOGRAPHIC ASPECTS OF SCHOOL ENROLMENT: THE CASE OF PRIMARY SCHOOLING IN ETHIOPIA^·

Mulugeta Gebreselassie and Amanuel Gebru 

Abstract: Primary education makes people literate and numerate thereby enabling them to effectively manage their domestic and occupational duties. In Ethiopia the rate of primary school participation is very low even when compared with other Sub-Saharan African countries. Officially, the enrolment rate is 34% while, according to this study, it is 36% (with standard deviation 0.004). Based on the low levels of enrolment this study looks into the salient socio-economic and demographic aspects of school enrolment in Ethiopia. It uses the logistic regression method of analysis among several methods that can be used to predict a binary dependent variable from a set of explanatory variables. The results show that the probability of primary school enrolment is a function of variables which are related to the family and to the child. In fact variables related to the location of residence of the child and their interaction with the variable related to the family and the child are also found significant.

Key Phrases: Binary dependent, logistic regression, primary schooling.

1. INTRODUCTION

Primary education equips a population with literacy and numeracy skills to meet the challenges of life at home and at work, laying a foundation for further educational advances (Lockheed and Verspoor 1991). It also provides the basis for developing the capacity to cope with rapidly evolving and changing societies in an information age. Its universal availability and quality are central to the human resource capacity of any society. The longitudinal studies of Bennet (1985, cited in Lockheed and Verspoor 1991) covering the period 1938-80 revealed that primary schooling had a considerable positive effect on the economic development of 110 Third World and First World countries.

Similarly Lau and Jamison (1991) showed that the economies of 22 Asian and Latin American countries were massively boosted by the beneficial effects of primary education. Also Lockheed and Verspoor (1980), in a review of 32 studies of rural areas, indicated that a farmer's education of four years boosted agricultural productivity by 8.7%. Similarly Psacharopoulos (1985) showed that educational investment yields significant increases in economic productivity. His longitudinal studies of 60 countries show that in developing nations the rate of return to primary education was 27% compared to 16% and 13% for secondary and higher education, respectively. Also the primary school experiences of a number of developing nations in the 1950s and 1960s were correlated with speeded up development in those countries in the 1960s and 1970s (Browman and Anderson 1973; Peaslee 1967). These imply the need for a maximum effort to achieve universal primary schooling.

Some African countries have already achieved and even gone beyond this universalisation of primary education. Remarkable examples with impressive enrolments include, Togo (132.7%), Swaziland (128.5%), Tunisia (114.4%), Algeria (107.3%) and Mauritius (106.7%). Promising results were achieved in Senegal (66.4%), Tanzania (55%) and Djibouti (37.3%) (Tegegne 1998). In the case of Ethiopia, though recent trends are encouraging, the percentage of school age children who participate in primary education is still very low. The 1996 education statistics revealed that Gross Primary Enrolment was 19.7% in 1992/93, 23% in 1993/94, 29% in 1994/95 and greater than 34% in 1995/96 (MOE 1997). This enrolment rate is less than sufficient for the country's requirements and development prospects. To have its population modestly involved in economic development, a country needs a minimum literacy rate of 40%, which is again less than sufficient for a meaningful economic growth (Browman and Anderson 1973). The message for Ethiopia is that primary schooling should be the most important priority for educational investment.

Lack of universal education has had dire consequences for Ethiopia: it is characterised by a high prevalence of preventable infectious diseases resulting in high rates of mortality and low life expectancy (World Bank 1999). Studies (Tilahun 1996) show that Ethiopia has infant and child mortality rates of 123 and 208 per 1000 live births - very high even by African standards. Incompatibly with its economy Ethiopia also has the third largest population in Africa. The population growth per annum of 3.3 percent could well be a predictor of an impending population crisis. For the period 1990-1995 the average fertility rate was a high 6.6 (Zelalem 1996). Ethiopia's contraceptive use of 7.5% compared with, for instance, Kenya's (23.2%) for 1989 and Zimbabwe's (32.2%) for 1988 is very low indeed (Zubedia 1992). The uncontrolled population growth against the nation's adverse economic realities, viz., low agricultural productivity, recurrent famine, macro-economy, low GNP per person (USD 110) (ranked as the 132nd lowest in the world; World Bank 1999) is a cause for concern and calls for intervention. As might be expected, Ethiopia also has one of the lowest levels of human development in Africa as well as the lowest rate of primary school participation.

A number of reasons have been presented as explaining the low level of school enrolment. Destafano and Wilder (1992) argue that dissatisfaction with declining school standards is a major reason for falling enrolments. In an earlier research, Destafano and Wilder (1992) mention the absence of sufficient opportunities for higher level education and the attendant unrealisable promises for a non-agrarian occupation, rising costs of schooling, disparity between received schooling realities of rural life and compulsory education as contributory to low enrolments.

A more comprehensive report on school demand in rural Ethiopia is USAID (1993), which involved a total of 520 households with school-age children from Bale, Welayita, South Gondar and Central Tigray in Ethiopia. The report cites economic difficulties as the most significant factors constricting primary school attendance and persistence in rural areas. The study indicates that direct and indirect costs were the most important impediments affecting parents' decision to send their children to school. Those who reported sending their sons to school did so in the belief that boys' education was a worthwhile investment in the long term. Woodhall and Psacharopoulos (1995) found a link between gender differences and values and attitude as determinants of school demand. The study also found that income represented by a house with a corrugated iron roof is positively correlated with schooling demand. Also the study found that in large families there is a smaller demand to send an additional child to school, if others are already enrolled. Understandably some children may have to work on the family farm as other siblings go to school.

The Policy for Human Resources Development (PHRD 1996) Report mentions the chief impediments to school enrolment in Ethiopia to be school distance and opportunity costs of boys' and girls' time. It also mentions the role of academic difficulties, deficient financial position, and poor quality education in determining household school demand in Ethiopia.

On the basis of similar data as the PHRD, Mulugeta, (1998a and b) advocates the education of parents, especially the education of the mother, to alleviate the problem of low school demand in Ethiopia. Though the studies tried to build models for each level of schooling (primary, secondary and overall) they emphasise the factors affecting overall schooling in Ethiopia. They have also excluded some important factors such as age-grade distortion, which can easily be derived from the available data.

Similar studies on other developing countries have produced similar results. The Sudanese study of Maglad (1994) found for instance that the most significant determinants of school attendance were age and sex of child, parental level of education, direct and indirect costs of schooling and region. In a similar Brazilian study, Singh (1992) indicated that family-size was another significant variable influencing enrolment. Also in a similar Malawi study on the determinants of the proportion of eligible siblings enrolled, Tan, Lee and Mingat (1984) found that direct school expenses affected negatively statistically significantly the number of school-age children that households can send to school. Mother's education, urban/rural residence and the proportion of daughters in the family were also significant determinants.

The proportion of children in school was higher in households with better-educated mothers, and the effect is reinforced in urban households. Father's income and residence in the northern region were found to be positively associated with high enrolment, where the latter possibly reflects the educational tradition in this part of the country. As might be expected, the proportion of girls was found to be negatively associated with enrolment. The study also found that the proportion of children sent to school was higher in urban households. Paternal earnings and living in the northern region were positively linked with higher school participation. The latter factor probably indicates an earlier introduction of education into this part of the country. Predictably the proportion of girls in a family was negatively correlated with enrolment.

Birdsall (1987) studied the demand for primary schooling in rural Mali. School fees, distance to school and school quality measured by student teacher ratio, number of books per classroom and delays in the payment of salaries to teachers had a negative association with enrolment. Among the household variables, the effect of income appeared to be negative when the ratio of adults to children was included as one of the explanatory variables. This may reflect the contribution to income of children not in school. Ethnic and religious differences across households had no effect on demand once other factors were taken into account. The study also found that school fee is more negatively correlated with demand than distance. A study of rural Peru (Gertler and Glewwe 1989) found level of parents' education, presence of other children of school-age, sex, school quality and cost of schooling to be the most important determinants of schooling.

A more comprehensive work by Lindsey (1994) carried out for the Ministry of National Education of Madagascar and UNESCO is another relevant study. Its principal objective was to identify factors influencing the admission to, attendance at, and dropping out of school in the primary school level in Madagascar. The study first classified the explanatory variables as characteristics of the child, the family, the school and the village. In the section dealing with enrolment the study used a meticulous model-building strategy which could serve as a reference for other similar studies. It concluded that the main determinants of primary school enrolment in Madagascar were factors which were related to the child followed by factors related to the family. Age and sex of the child and education level and religion of parents were found to be especially crucial variables.

However, unlike studies on other countries, empirical studies related to determinants of primary school enrolment in Ethiopia are quite limited. Accordingly, the aim of this study is to examine the effect of salient socio-economic and demographic factors on primary school enrolment in Ethiopia.

2. OVERVIEW OF METHODOLOGY

2.1 The Survey

The survey covers all regions in the country according to the principles of multistage probability sampling (CSA 1997). The sampling unit was a household. About 11000 households were studied and about 57000 school age persons were registered of whom about 13770 were primary school age persons. The quality of the data is reliable because the survey was carried out by the Central Statistics Authority (CSA) which undertakes national surveys and censuses by using its professional employees and by training additional personnel. In fact CSA is a national institution which is authorised to do for survey design and data collection.

2.2 Variables

The survey covers about 500 socio-economic and demographic variables. Among these about 30 are considered in this study. The dependent variable designated as P_ENROL is defined as a discrete binary variable which assumes the value 1 if a sampled primary school-age person (ages 7-14) is enrolled in grades 1 to 8 or 0 if not. And the explanatory variables, which are mainly related to the characteristics of the child, his/her parents, the community and the region (state), are defined as follows:

1. Age in completed years (AGE);

2. Age-grade distortion (AGEDIST) coded as 1 if the person is enrolled in the appropriate grade (for instance grade 1 at age 7) and 0 if not;

3. Sex of the person (SEX) coded as 1 if male and as 0 if female.

4. Relation of the child to head and spouse (MUTCHLD) coded as 1 if mutual child and 2 if the child is only related to either of them, and 3 if any other relative;

5. Health status of the person (HEALTH) coded as 1 if the person is not disabled and 0 if he/she is disabled;

6. Marital status of the person (MAR_ST) coded as 1 if he/she is married and 0 if he/she has never been married;

7. Family Size (FAM_SIZE): the number of family members who at least cook and wash clothes together;

8. Health status of the head of the household (HLTH_HD) coded as 1 if the head is not disabled and 0 if he/she is disabled;

9. Age of the head of the household in completed years (AGE_HD);

10. Literacy status of the head of the household (LIT-HD) coded as 1 if the head is literate and 0 if he/she is illiterate;

11. Literacy status of spouse (LIT_SPS) coded as 1 if the spouse is literate and 0 if he/she is illiterate;

12. Sex of the head of the household (SEX_HEAD) coded as 1 if the head is male and 0 if she is female;

13. The natural logarithm of expenditure for schooling (LN_SCHEXP);

14. The natural logarithm of total expenditure (LN_TEXP);

15. Proportion of school-age girls within a household (PRP_GIRL);

16. Mother's education in years of schooling (MOTH_EDU);

17. Father's education in years of schooling (FATH_EDU);

18. Income of the household in Birr (INCOME);

19. Occupation of the head of the household (OCUP_HD) or spouse (OCUP_SPS) coded as 1 if the person is a farmer and 0 if not;

20. Employment status of head of the household (EMPL_HD) or spouse (EMPL_SPS) coded as 1 if he/she is currently working as an employee and 0 if not;

21. Number of school age people, persons of age 7-24 (FS724);

22. Distance to Primary School in KMS (DST_PRI);

23. Distance to Secondary School in KMS (DST_SEC);

24. Location of residence (URB_RUR) coded as 1 if the household resides in urban quarters and 0 if it resides in rural quarters;

25. The regional dummies assume the value 1 if the person under study is from the respective region and 0 if he/she is not from the region. The variables are coded as TIGRAY, AFAR, AMHARA, OROMIA, SOMALI, BEN-GUM, GAMBELA, HARARI, DIREDAWA, and ADDIS.

26. There are also interaction variables which are important to detect the interaction effect of some of the factors. They are identified by their size of common effect in enrolment when cross-tabulated to each other.

i. INT1: the interaction of SEX and URB_RUR

ii. INT2: the interaction of MUTCHLD and URB_RUR

iii. INT3: the interaction of MUTCHLD and SEX

iv. INT4: the interaction of URB_RUR and DST_PRI

v. INT5: the interaction of SEX and DST_PRI

vi. INT6: the interaction of MOTH_EDU and SEX.

2.3 Modelling

A variety of statistical models can be used to predict a binary dependent variable from a set of independent variables. But each of them has pros and cons. This study uses the logistic regression model, which requires fewer assumptions than the others do (Aldrich and Nelson 1984; John and Forrest, 1984; Agresti, 1996; Hosmer and Lemeshow 1989).

The process of model-building is somehow adapted from Lindsey (1994). Firstly all variables were entered into the model and those whose p-value was less than or equal to 5% were selected. Secondly progressive model-building was continued by first inserting the variables related to the family and the community (past studies showed these variables to be relatively important in determining primary schooling in Ethiopia) into the model and then those related to the child and the region after the variables which were not significant were avoided at each stage. A variable is considered significant if it reduces the deviance by about two units according to the Akaike Information Criterion. Furthermore, model validation was done by using Cox and Snell's R², and classification statistics. SPSS for WINDOWS version 9.0 was used for data analysis.

Finally the effect of missingness was investigated by defining covariates to the variables with many missing values because the problem of missingness distorts the parameters in the model (Lindsey and Lindsey 1999). It is known that the problem of missingness is a common phenomenon in surveys undertaken in developing countries because some people may not be contactable or they may refuse to participate and those replying may not answer all questions.

3. Results

3.1 Descriptive Results

The study covers 59% of the rural population and 41% of the urban population. These are shared among the regions as 6.4, 1.3, 21.7, 34.1, 0.8, 1.7, 16.0, 1.3, 2.7, 11.0, and 2.9 percent for Tigray, Afar, Amhara, Oromiya, Somali, Benishangul-Gumuz, SNNP, Gambella, Harari, Addis Ababa, and Dire Dawa, respectively.

The national rate of school enrolment observed in the study, 35.8%, is similar to that of the reported official national level. When seen across the regions, Addis Ababa, Harari, Dire Dawa and Tigray were found to outperform the other regions. Primary school enrolment was found worse in Afar, SNNP and Benshangul-Gumuz regions. It was found that the sex representation in the data is also satisfactory, about 50.1% of the candidates being females. From these about 58.4 % have never been married while 2% of them were involved in marriage. However, about 39.6% of them did not reply to the question asking marital status. The present grade of the students was also recorded and it was found that about 1% of the primary school-age persons were enrolled in secondary school grades. Some of the non-enrolled children had already completed some grades before they dropped out. And in those who completed some grades and who were not in school at the date of the survey a severe age-grade distortion was observed. Only 8.1% of the students were found in the appropriate age- grade composition. Overall 91.9% rate of age-grade distortion was observed in the study. When we look at the health conditions of the candidates about 89.9% of them were without any disability problem.

Similarly, when we look at the variables related to the family, it was found that 75.3% of the households were headed by males and the rest by females whose age ranged between 15 and 97. Among these (all household heads) about 63.7% were illiterate while 2% had university/college education. About 54.8% of the heads were farmers. Of these 27% had disability problems. And 19.3% lived without wife/husband (not married, divorced, separated or widowed). A discrepancy between the two groups was also observed in the education of parents measured in years of schooling. The average grade of the father and mother of enrolled children was found to be 8 and 4 while it was 5 and 1 for non-enrolled, respectively. A negative relationship was observed between family size and school enrolment. The average family size for enrolled children was 1 while it was 5 for non-enrolled children. Contrary to this, welfare status (measured by income and expenditure for schooling) was better for enrolled children than for non-enrolled ones. The average income and school expenditure of enrolled children were 1736.9 and 590 Birr, respectively, while they were 293 and 81.2 Birr for non-enrolled children.

The average distance between home and primary school was 1 km for those enrolled while it was 4 kms for those not enrolled. Similarly the distance to secondary school was 23:7 km for enrolled and non-enrolled, respectively.

One of the most important variables with a severe problem of missingness was education of parents measured in years of schooling. It was found that 1.7% of the fathers were illiterate but about 23.8% had some education. The remaining 74.5% did not reply to the question. The figures are similar for the mother except that illiterate mothers account for 16.3% of those who reported their grades. Similarly the variables family size and size of school-age people had 39.5% and 39.7% missing cases, respectively. It was observed that the family size of the households ranged between 1 and 16. Additional tables which describe the data are presented in the annexes.

3.2 The Models

With the purpose of showing the general feature of the predictors of primary school enrolment, preliminary results from a chi-square analysis are presented in table 1. As can be seen from the table, almost all the predictors are significantly associated with P_ENROL at less than one percent significance level. This implies the predictors are valid for further investigation.

Next to the preliminary assessment, further analyses were made to determine certain factors that can predict P_ENROL by using logistic regression. The predictors selected by the regression analysis were those that make the observed values of P_ENROL most likely. Estimates of the logistic regression model are presented in tables 2 to 7. The tables contain the estimated coefficients and related statistics. From table 2 it can be seen that most of the variables related to the child are significant except health of the child. The most important child related variables affecting school enrolment were AGEDIST, MUTCHLD, SEX, AGE and MAR_STA in that order. Especially being at the right grade for a given age (AGEDIST) was highly positively associated with school enrolment. This could be a manifestation of the phenomenon that those who completed some grades but who were not registered at the time of the survey were not at the right grade for a given age, while those who were enrolled were somehow better off. In fact as tables 5a and 5b (see Annex) show, the age-grade distortion was wide even among those enrolled. Only 22, 19.7, 16.3, 3.7, 20.1, 28.7, 31.5 and 51.3 percent of those registered at a given grade were in their right age for the consecutive primary grades. From the data it can also be seen that only 1.1% of the rural primary school-age people and only 18.1% of the urban ones attend at the right grade for a given age.

Table 1. Pearson chi-square test

*** p<0.01 and * p<0.05.

As expected gender differences also significantly affected enrolment. A male was more than twice as likely to be enrolled as a female. Similarly mutual children (children of both the household head and the spouse) had a better chance of going to primary school than non-mutual children did. Children of either the head or the spouse had a better chance of going to school than other relatives.

The model in table 2 seems to satisfy loosely the criterion for model adequacy. The Cox Snell coefficient of determination shows that only about 14% of the variability in the probability of school enrolment is explained by the child related variables. Similarly the classification statistics which shows a 64.73% correct classification power is satisfactory.

Table 2. School enrolment with variables for the children (N= 13770, M= 5497)

Table 3 shows the logistic regression results of those variables related to the family when SEX and AGEDIST are not removed from the model. Similar to the above model the effect of AGEDIST was very significant. Similarly children of literate parents were found to have more chance of going to school than those with illiterate parents. Being a farmer's child, which is strongly associated with having an illiterate parent, was also found to be a disadvantage. Older parents seem to support the enrolment of their children. The effect of family size was negative while the effect of having a large number of school-age children was positive.

Table 3. School enrolment with variables related to the family (N= 13770,

M=10257)

Significant Variable	Estimate	Standard Error	Omitted Variable
SEX	.4648	.0974	INCOMHD
AGEDIST	8.6384	4.0769	HEALTHHD
AGE_HD	.0487	.0047	EMPL_HD
LIT_HD	2.2051	.2449	EMPL_SPS
LIT_SPS	2.3122	.3001	SCH_EXP
FATH_EDU	.1111	.0153
MOTH_EDU	-.0267	.0136
FAM_SIZE	-.1720	.0460
FS724	.2141	.0733
PRP_GIRL	-.6536	.2644
OCUP_HD	-1.5054	.1538
OCUP_SPS	-1.8702	.1872
Constant	-4.1947	.3513

In the model shown in table 3, the number of missing cases is very large. This came as a result of the variables MOTH-EDU and FATH_EDU. Nevertheless the values of Cox Snell's R² and the classification power are by far better than the model in table 1(R² = 42% and classification =81.53%). This reveals the better predictive condition of family related variables.

The regional variables were also found significant except for AFAR and BEN-GUM. It is apparent from table 4 that residing in Amhara, Oromiya, Somali and Dire Dawa regions reduced the chance of going to school. The variables SEX and AGEDIST were also highly significant as in the tables 2 and 3. Distance to both primary and secondary school was also negatively related with school enrolment. The negative effect of distant primary schools is evident but for that of distant secondary schools the idea is that it discourages both parents and children of primary school-age not to pursue their primary education for they take it as an overwhelming obstacle in the way of the pursuance of secondary education. The positive effect of living in urban areas, which is obvious, is also confirmed in this study.

Model validation parameters are also significant. The Cox Snell's coefficient of determination shows that about 39.3 percent of the variability in school enrolment is explained by the regional and community variables. In fact the classification power, 81.94%, is also quite high.

Table 4. School enrolment with variables related to the community (N= 13770, M= 1)

When all the variables are taken together, it was found that AGE, AGEDIST, DIST_PRI, DIST_SEC as well as regional variables such as ADDIS, and AFAR were not significant. But SEX was consistently positive and significant. Similarly the positive effect of being a mutual child of the head and the spouse was significant in table 5.

The percentage of variability in the likelihood of enrolment is maximised in this model. About 46% of the variability is explained by the variables therein. In fact the model classifies about 84.92% of the cases correctly.

Table 5. School enrolment with variables related to children, the

family and the community (N= 13770, M= 11589)

Significant Variable	Estimate	Standard Error	Omitted Variable
SEX	.5315	.1387	AGE
MUTCHLD			HEALTH
MUTUAL(1)	-.8110	.2098	MAR_ST
MUTUAL(2)	-2.5316	.1770	AGEDIST
AGE_HD	.0185	.0065	INCOME
LIT_HD	2.4797	.3105	HEALTH_HD
LIT_SPS	3.0111	.2901	PRP_GIRL
FATH_EDU	.1490	.0227	DST_PRI
MOTH_EDU	-.0671	.0196	DST_SEC
FAM_SIZE	-.1785	.0770	ADDIS
FS724	.2726	.1042	AFAR
OCUP_HD	-1.1803	.2522	SOMALI
OCUP_SPS	-1.3202	.2786	BEN_GUM
URB_RUR	1.2969	.4307	GAMBELA
TIGRAY	1.2423	.4385	HARARI
AMHARA	-.6332	.2466	DIRE DAWA
OROMIYA	-.4282	.2103	EMPL_HD
Constant	-3.7350	.8542	EMPL_SPS

The interaction effect of some of the variables together with some other variables was also investigated in table 6. The coefficient of MUTCHLD was negative to show that mutual children have more chance of school enrolment than the other two groups. Literacy status of parents was still positively related with school enrolment. The interaction variables INT1, INT2 and INT4 were significant. The coefficient of INT1 was negative showing the reinforced negative effect of being a rural girl on her chances of going to school. The coefficient of INT2 revealed the positive relationship between urban residing relatives (non- mutual children) and school enrolment, which could manifest the rural to urban migration of children in search for education. The positive coefficient in INT4 shows the positive association between living in rural areas and long distances and ultimately their joint negative effect on enrolment. Long distance schools seem to be positively correlated with higher chances of enrolment.

The percentage of variability in the likelihood of enrolment which is explained by the model in table 6 is also maximum. About 44% of the variability is explained by the variables in the model. In fact the model classifies about 82.5% of the cases correctly.

Table 6. School enrolment with variables related to children, the family and the

community and some of their interactions

Table 7 shows significant variables selected by a backward elimination procedure of SPSSWIN after 20 iterations. The results are closely similar to that of table 6 except for a slight difference in the values of the coefficient of determination and power of classification.

Table 7. School enrolment with variables related to children,

the family and the community which are selected by

backward elimination (N=13770, M=9272)

Finally attempts were made to study the effect of non-random missingness on the models constructed. It was found that, though there were several missing cases on some of the models due to some variables, inserting the covariate form of the variables did not differ significantly from that of the normal entries. This may be an indication of the randomness of the missing cases which may also result from the large sample size.

4. DISCUSSION, CONCLUSION AND RECOMMENDATIONS

The results indicate that the probability of going to school is a function of variables related to the family, the region, the community and the child. These groups of variables explained 39.3, 42, and 14 percent of the variability in enrolment. Interaction of some the variables is also found important.

Parental education, measured as the literacy status of the parents and in the number of years of schooling received, influences school enrolment positively. Educated household heads that are government employees favour the enrolment of their children as they have experienced occupationally the liberating consequences of education. Similar results are found as concerns parents' education, measured in years of schooling, which likewise is a positive correlate of enrolment. Together the results show that educating parents, as a strategy, can increase school participation rates.

Occupation of parents also has a significant influence on the probability of school enrolment. Farmers (or in a general sense self-employed parents) are found to be less likely to send their children to school. The reason could be the higher need for boys'/girls' labour by farmers than government employees, who are less likely to need their children's time but more likely to appreciate the returns from education. Of course the negative effect of being self-employed, especially in literate parents, is reinforced in rural males. It may be expected that this problem will be solved in due course with the creation of non-farming jobs in the rural areas of Ethiopia. The developments observed in this line after the war years are in evidence and as such a source of optimism in a greater generation of non-agrarian jobs.

The link between school enrolment and age of the head was found to be positive. Manifestly older parents were found to be sending their children to school as a forward-looking investment because they feel that they will be helped by their children's income later, perhaps in old age. The returns to education are probably more pertinent to low-income families with children at school. Such families may view parental investment in education as a source of supplementary income to their inadequate wages and to those who take the long view as an insurance against insufficient retirement benefits. Beyond this there are psychological returns when ones' children are educated, successful and independent, and in the case of rural parents, when their children have non-agrarian occupations, which marks for them an occupational liberation.

With respect to distance the results lend confirmatory support to previous studies demonstrating distance as a deterrent factor as in Rose et al. (1996). We found that an additional kilo meter of distance to primary school decreases school enrolment by 7 percent while distance to secondary school decreases by 0.4%. The effect of distance, as expected, is higher in rural areas, and differentially so for rural females. This could be for the reason that distance to school is associated with temporal expenditure, i.e., high costs. That is, when the school is outside the households' residence, the student will be forced to stay out of his/her village for the whole school season. This has big direct and indirect cost implications. The household will have to expend additionally for transport and food, in addition to the time spent in travelling. For females, especially for those at the puberty stage, the attendant worry of the parents can have a significant negative effect on their enrolment. In addition the negative influence observed in distance to secondary school could be seen from the point of the negative psychological impact it could have on parents and children in sending their children or going to primary school. The absence of a secondary school deters primary school enrolment because those intending to go to primary school are frustrated ahead of time about the absence of opportunities for completing secondary education. This is for the simple reason that parents do not develop a sense of fulfilment unless their children complete secondary education. They consider completing primary education and not moving further as wastage because of the low returns observed from primary education in Ethiopia to date.

However in urban areas and particularly in the Metropolis, rich households favor distant schools in search of quality education in private schools. Understandably, however, these represent a tiny segment of the Ethiopian population constantly struggling to make ends meet. For the vast majority, as the results indicate, distance is a significant deterrent. Thus, if this negative impact on school enrolment is to be controlled the government should emphasise constructing more schools in rural areas of Ethiopia or involve massively Ethiopian and international NGOs in achieving this goal. Successful involvement of NGOs in primary schooling was witnessed in Pakistan, Bangladesh and India (King and Hill 1993). Financial incentives have been offered in a number of countries. Bangladesh and Malawi for instance offer fee waivers for nonrepeaters (Stromguist 1997). In Ethiopia it may be quite important that the government encourage private investment in primary education which may ease the burden of state schools.

The results further demonstrate the presence of parental bias in favor of their mutual children. Relatively speaking parents do not give priority to the education of their relatives as well as their exclusive children in the presence of mutual children. This is particularly observed in rural areas. In urban areas usually children migrate from rural areas for the sake of education and the question of priority is not a problem. However, this is not as significant as sex bias overall. As a reflection perhaps of the country's low level of development and the attendant cultural attitude to females and their education which adversely affects their aspiration levels, boys' enrolment probability is found to be more than twice that of girls. Therefore, to consolidate the increasing trend of school participation in the last five years, the government should devise gender sensitive policies and place special emphasis on promoting female enrolment. Intervention programs that target girls may prove to be important (Bouis 1998). Todays' girls are tomorrow's mothers and it is well known that educated mothers are more likely to send their daughters to school. There is a positive relationship between women's education and male life expectancy and countries with the smallest educational gaps between males and females have higher GNPs (King and Hill 1993).

The results confirm the findings of Teklehaimanot (1999) about the considerable problem of age distortion in urban and more particularly in rural areas. While 18.1 % of urban children are sent to school at the right age, only 1.1% of rural school children were in a grade level appropriate for their age. If at all rural parents send their children to school, it is usually when the children are overage for a grade level. Parents seem to perceive that children at 7 are too young for school. Rural children specially are usually undernourished and their growth stunted, distorting a correct correspondence between their physical appearance and age. The distortion means that rural children will remain in school when they should have been at work as graduates.

Similarly, school participation rates appear to vary by region and location of residence. As expected, participation rates in urban areas are found to be higher than in rural areas. But the magnitude varies from region to region. For instance the likelihood of going to school in Tigray region, which has a long history of church education and therefore higher levels of literacy, seems greater than that of the Southern Nations, Nationalities and Peoples' Region (SNNPR) which is also less well urbanised. Also, especially after the war years, there has been considerable mass sensitisation and NGO-led construction and reconstruction of schools in this potentially less rich region. On the other hand, the likelihood of going to school in SNNPR is greater than in the Amhara, Oromiya, Somali and Dire Dawa regions. This may explain the vast missionisation of the Southern Region, which has led to a number of schools opening up and this has had a multiplier effect. In contrast, in the Somali Region, enrolment is considerably lower, which may be explained by the fact that this region is one of the least urbanised, partly because of the nomadic lifestyle of the local population, and possibly due to a systematic official neglect in the past.

The presence of another school-age person in the household is found to increase the probability of school enrolment. This could be attributed to many factors. However, in Ethiopia, especially in the rural areas, double-shift systems have not been introduced. That means parents do not have the chance to send their children to school alternatively, as the demand for children's labour is very high in agrarian communities. The effect is indeed aggravated during the main agricultural seasons. Therefore, more should be done to alleviate the problem. Introducing double-shift systems and adjusting the school schedule, when necessary, to less hectic seasons could be one way to alleviate the problem. Contrary to this, large families were found associated with low enrolment. The fact that the population of Ethiopia is mostly very young and also the considerable number of old age dependants in Ethiopia could contribute to this effect.

The interaction of some of the variables has also a significant influence on the probability of school enrolment. The interaction of sex and location of residence, inserted in the equation to capture the reinforced effect of living in rural areas on females, is found to be negatively associated with enrolment. Similarly the interactions of MUTCHLD and DST_PRI with location of residence were found significant. It was found that living in rural areas, which is associated with illiterate parents, distant schools and low socio-economic status reduces the chance of going to school. Hence, urbanisation in the sense of educated parents, more schools and better socio-economic conditions could improve school enrolment. Altogether educating parents, as a strategy can increase school participation rates. However, if the enrolment rates remain low there are various policy implications.

Firstly the government's education policies may not be realisable. The new government has as its goal the universalisation of primary education by 2008. The allocation of substantial funds towards the realisation of this goal would be denied meaning. This is quite harmful in view of the fact that there is a fierce budgetary competition among sectors for official attention. The costs of the training of teachers, preparation of texts and construction of schools; indeed all state educational expenditure earmarked for this purposes would be invalidated by non-enrolment. Viewed against the background that primary education is free, the practical consequences of non-enrolment are considerable indeed.

Demographically, non-enrolment and the attendant lack of schooling would make ineffectual the country's population policy. Research shows that women's education is a strong predictor of fertility and contraceptive use. Thus undereducated mothers are likely to be prolific and unreceptive to family planning interventions. The practical consequences of unplanned population growth for Ethiopia are capable of hampering its development endeavours.

Significant figures of non-enrolment would also have implications for Ethiopia's agricultural policy to modernise farming practices and boost production. Uneducated farmers do not produce enough even for themselves because they operate under ignorance and are bound by tradition. The fact that Ethiopia has rarely attained food self-sufficiency while ironically over 85% of its population is agrarian speaks for itself. Education would help produce `worker and allocative effects' (Welch 1978). `Worker effect' refers to the incremental change in production owing to a unit change in education, i.e., the additional schooling equips a worker with increased productive competence. The allocative effect on the other hand refers to the `capacity to evaluate and adopt profitable new technologies' (Cotlear 1989, 75). This includes receptivity to modern agricultural practices and extension programs, which can lead to higher levels of agricultural productivity. Lockheed and Verspoor (1980) cite a World Bank study of 18 countries which showed that primary education led to an increase in productivity by 13%. It issued from the use of new seeds, fertilisers and raised farmers' income. Similar evidence comes from Korea, Malaysia and Thailand (Jamison and Lau 1982) and India, Bangladesh and Pakistan (Jamison and Mockle 1984).

The implications of non-enrolment will also weigh more heavily on women and make inoperable gender policies of the country to reverse the inequalities of the past. If women do not enrol equally with men, the consequences thereof are disproportionately higher for them. They are more likely to remain unemployable. In 1988, only 18% were employed (Rose et al. 1996). Even at present, rates of unemployment for them are far higher. In consequence of non-enrolment, policies designed to empower women are rendered unworkable. This has far-reaching consequences. Recently, a multi-country study on the benefits of girls' education found that nations that attained universal primary education for males in 1965 but in which girls' participation lagged behind had double infant mortality and fertility rates 20 years later than those countries in which gender equity was attained (Snyder and Taddesse 1995).

Overall non-enrolment is likely to impede the development policies of the country, as an undereducated population would not be able to engage meaningfully and productively in national economic development. It would imply perpetual underdevelopment, poverty and dependence on foreign aid, and recurrent conflict. This has its own implications for nation building, which does not come so easily in the absence of a civil society with a degree of political consensus.

REFERENCES

Agresti, A. 1996. An introduction to categorical data analysis. New York: John Wiley.

Aldrich, H., and D. Nelson. 1984. Linear Probability, Logit and Probit Models. Newbury Park: Sage Publications.

Birdsall, N. 1987. Demand for primary schooling in rural Mali: Should user fees be increased? World Bank.

Bouis, H. et al. 1998. Gender equality and investment in adolescents in the rural Philippines. Research Report No. 108. International Food Research Institute.

Browman, M. J., and A. C. Anderson. 1963. The role of education in development. In Old societies and new states: The quest for modernity in Asia and Africa, edited by C. Geertz. New York: Free Press.

_____. 1973. Human capital and economic modernization in historical perspective. In Fourth International Conference of Economic History, edited by Lane F. C. Paris: Mount De Grunter.

Central Statistics Authority (CSA). 1997. Household income, consumption and expenditure survey report. Addis Ababa.

Cotlear, D. 1989. The effects of education on farm productivity. Journal of Development Planning, no.19: 73-99.

Destafano, J., and A. Wilder. 1992. Ethiopian Education Sector Review. USAID.

_____. 1993. The demand for primary schooling in rural Ethiopia. Addis Ababa, USAID.

Gertler, P., and P. Glewwe. 1989. The willingness to pay for education in developing countries. Washington, D.C.: World Bank.

Hosmer, W., and S. Lemeshow. 1989. Applied logistic regression. New York: John Wiley.

Jamison, D., and L. Lau. 1982. Farmer education and farm efficiency. Baltimore: Johns Hopkins University.

Jamison, D., and P. Mockle. 1984. Farmer education and farm efficiency in Nepal: The role of schooling, extension services and cognitive skills. World Development 12, no. 1: 67-86.

John, H. A., and D. N. Forrest. 1984. Linear Probability, Logit, and Probit Models. Newbory Park: Sage Publications.

King, E., and A. Hill, eds. 1993. Women's education in developing countries: Barriers, benefits and policies. Baltimore and London: Johns Hopkins University.

Lau, Lawrence, and Jamison Dean. 1991. Education and productivity in developing countries: An aggregate production function approach. Policy, Research and External Affairs Working Paper 612. Washington, D.C.: World Bank.

Lindsey, J. K. 1994. Survey report of primary education in Madagascar. http://www.luc.be.~jlindsey.html/.

Lindsey, J. K., and J. P. Lindsey. 1999. Detecting covariates with nonrandom missing values in a survey of primary education in Madagascar. http://www.luc.be.~jlindsey.html/.

Lockheed, M., and A. M. Verspoor. 1980. Farmer calculation and farm efficiency: A survey. In Education and income: A background study for world development report 1980, edited by T. King. World Bank Staff Working Paper No. 402. Washington, D.C.: World Bank.

_____. 1991. Improving primary education in developing countries. New York: Oxford University Press.

Maglad, N. E. A. 1994. School supply, family background and gender specific school enrolment and attainment in Sudan. Eastern Africa Social Science Research Review 10, no. 2: 1-20.

Ministry of Education (MOE). 1997. Education statistics. Addis Ababa.

Mulugeta Gebreselassie. 1998a. Determinants of household school demand in Ethiopia: A multivariate assessment. Eastern Africa Social Science Research Review 14, no. 2.

_____. 1998b. Determinants of school enrolment in Ethiopia. In Proceeding of the conference on the quality of primary and secondary schooling in Ethiopia, edited by Amare Asgedom. Addis Ababa.

Peaslee, A. 1967. Primary school enrolment and economic growth. Comparative Education Review 11, no. 1:57-68.

PHRD. 1996. Household demand for schooling. Addis Ababa.

Psacharopoulos, G. 1981. Returns to education: An updated international comparison. Comparative Education 17: 321-314.

_____. 1985. Returns to education: A further international update and implications. Journal of Human Resources, 20: 583-604.

Rose, Pauline, Getachew Yoseph, Asmaru Berihun, and Tegegne Nuresu. 1996. Gender and primary schooling in Ethiopia. Sussex: Institute of Development Research.

Singh, R. D. 1992. Under-investment, low economic returns to education, and the schooling of rural children: Some evidence from Brazil. Economic Development and Cultural Change 40, no.3: 645 - 664.

Stromguist, N. 1997. Increasing girls' and women's participation in basic education. Paris: International Institute for Educational Planning, UNESCO.

Synder, M. C., and M. Taddesse. 1995. African women and development. Johannesburg: Witwatersrand University Press.

Tan, J., K. Lee, and A. Mingat. 1984. User charges for education: The ability and willingness to pay in Malawi. World Bank Staff Paper No. 661. Washington, D.C.: World Bank.

Tegegne Nuresu. 1998. Indicators of access and wastage in the primary schools of Ethiopia. Proceedings of the conference on the quality of primary and secondary schooling in Ethiopia, edited by Amare Asgedom. Addis Ababa.

Tilahun Teka. 1996. Causes of child mortality in Gondar Hospital, Ethiopia. Ethiopian Journal of Health Development 10, no. 3: 145-148.

Teklehaimanot Haileselassie. 1999. Characteristics and implications of age-grade distortion in the Ethiopian primary education. IER FLAMBUE 6, no. 2.

USAID. 1993. Demand for schooling in rural Ethiopia. Addis Ababa.

Welch, F. 1978. The role of investments in human capital in agriculture. In Distortions of agricultural incentives, edited by T. W. Schultz. Bloomington: Indiana University Press.

Woodhall, M., and G. Psacharopoulos. 1995. Education for development: An analysis of investment choice. 5th edition. N.Y.: Oxford University Press, World Bank.

World Bank. 1999. World Development Report. Oxfor Oxford University Press.

Zelalem Fekadu. 1996. Social-psychological factors associated with contraceptive use: Attitudes of married women in the Kechene community of Addis Ababa. Ethiopian Journal of Health Development 10, no. 3: 153-160.

ANNEX

Table 1. Regional comparison of primary school enrolment

Table 2. Percentage of enrolment for each category of variables related to the child, the family and the community

Table 3. Descriptive statistics

Table 4. Age-grade distortion: From survey data

Table 5. Age-grade distortion: From MOE Statistics Abstracts Data

				Grade
Age	1	2	3	4	5	6	7	8
7	0.22	0.043	0.015	0.0097	0.013	0.014
8	0.25	0.185	0.078	0.0270	0.010	0.010
9	0.17	0.219	0.193	0.1117	0.040	0.014
10	0.18	0.234	0.269	0.2685	0.193	0.084
11	0.08	0.175	0.227	0.2879	0.341	0.308
12	0.09	0.144	0.218	0.2952	0.404	0.570
13							0.505	0.355
14							0.495	0.645

1 Department of Statistics, Addis Ababa University. Address for correspondence: P.O. Box 34613, Addis Ababa, Ethiopia; e-mail: Mulugeta_g@hotmail.com

2 Department of Foreign Languages and Literature, Addis Ababa University.