Determinants of youth unemployment in Uganda: The role of gender, education, residence, and age

The data for this study are obtained from the most recent Uganda National Household Survey collected from June 2016 to June 2017 by UBOS. During the survey, each household was interviewed twice a year, in visits 6 months apart. While in the first stage, 1,750 Enumeration areas (EA) were selected, in the second stage 10 households were randomly selected in each Enumeration area, bringing the total households to 17,320 - a sample size that more than doubled that of 2012/2013 of 6,887 households. Relative to all similar surveys before, the UNHS 2016/17 was thus nationally more representative. As evident in Table 1, a total sample of 5,912 respondents aged between 18 years and 30 years was retrieved from the main dataset of which 10.20% were unemployed during the time of the survey.

The use of a range up to 30 years is because in the Ugandan context, a youth in Uganda is a person who is aged 18–30 years (UBOS, 2017). Anyone <18 years is a child and engaging in work at that age is considered child labor in Uganda and is illegal. We therefore prefer the Ugandan contextual definition of the youth to that of the United Nations since we are looking at youth unemployment in Uganda.

Data was collected on gender/sex (as in Asalfew, 2011 – for Ethiopia); health (as in Hoorn, 2007); and education (as in Bhorat, 2007 - for South Africa); without sidelining other critical variables previously identified in the literature. Specifically, the latter include age (as in Mahlwele, 2012 – for South Africa); residence (as in Mulu, 2012 – for Ethiopia); region (as in Bhorat, 2007 – for South Africa; and, marital status (as in Msigwa and Kipesha, 2013 – for Tanzania). Education is coded “0” for no education, “1” for primary completion, “2” for secondary completion, and “3” for post-secondary completion; sex is a binary variable coded as “0” for females and “1” for male; health is also a binary variable coded “0” for no illness/injuries, and “1” for illness/injuries; residence is binary coded “0” for rural, and “1” for urban; region is coded “0” for central “1” for eastern, “2” for northern “3” for western; marital status is coded as “0” for single “1” for married, “2” for divorced/separated, and “3” for widow/widower; and, age of youth is taken as a continuous variable. The reference category used in the model were respectively No education (ref.), Female (ref.), Good health (No, ref.), Central (ref.), Rural (ref.), and, Single (ref.). We expect each of these variables to influence youth unemployment. A summary of the descriptive characteristics of the dependent variable (unemployment) together with the independent variables, viz., demographic and socio-economic factors, is presented in Table 1.

As observed in Table 1, the proportion of youth unemployment varied from one region to the other. While the highest percentage of unemployed youth was observed in Central (17.71%) followed by Northern (7.43%), the lowest was recorded in Eastern (6.57%) followed by Western (6.84%). Similarly, the proportion of youth unemployment differs by place of residence. Accordingly, a big number of unemployed youth (15.13%) resided in urban and relatively a small number of unemployed youth (5.90%) resided in rural areas. On the other hand, the proportion of youth unemployment varies by marital status. For example, the biggest percentage was observed in divorced youth (32.37%) followed by married ones (16.13%) and then the single youth (7.99%). Moreover, it is evident from the table that the proportion of unemployed youth differs by their age groups: the highest proportion of unemployment is observed in the age group 26–30 years (22.82%) and the lowest proportion in the age group 16–20 years (4.77%). It is interesting to further observe that statistics on health status show that the highest percentage of unemployment is found in youth with bad health (10.31%) whereas the lowest (10.10%) is notable for those in good health. Additionally, the female youth (10.44%) were more exposed to unemployment compared to their male counterparts (9.86%). Finally, we note that the unemployment status of the youth varies by their level of education. The highest percentage of youth unemployment is here observed in the youth with post-secondary education followed by those with secondary education whereas the least percentage is visible in those without education.

A further description of our data can be found in Table 2 exhibiting the correlation matrix. It is clear from the table that none of the correlation values exceeds 80%, pointing possibly to no multicollinearity issues as well as some possible relationship between the independent variables. However, although correlation coefficients measure the covariability of variables, they do not necessarily imply functional relationships between the variables. Therefore, we proceed to the regression analysis. But before we do that, there is a need to carry out diagnostic tests on our model. This follows in subsection 3.2.2.

An econometric requirement before estimating binary logistic regression model is to test for multicollinearity and the goodness of fit of the model. As reported in Garson (2009), multicollinearity in logistic regression is a result of strong inter-correlation among the predictor variables. Several measures can be used in assessing multicollinearity including but not limited to the Variance Inflation Factor (VIF) and Tolerance. The latter is captured by, 1–R² (coefficient of determination). As the value R² approaches 0, implying that the inter-correlation is lower, the estimate approaches 1.

On the other hand, VIF is the reciprocal of Tolerance expressed as 1/(1–R²). By rule of thumb, a VIF value equal or >4 (arbitrary but common cutoff criteria) implies that the independent variable exhibits multicollinearity (Hair et al., 2010). VIF measures how much of the variances are inflated, and can be computed either from the simple correlation coefficient or the partial correlation coefficient.

VIF = 1/(1−R²) for either Y = f(X) or Y = f(X₁,X₂)

If R² = 1, then VIF = ∞ implying perfect multicollinearity. Otherwise, if R² = 0, then VIF = 1 that means that there is no multi-collinearity. In essence, as the value of the coefficient of determination (R²) increases, the higher is the VIF, and thus there are increased chances or indications of multi-collinearity. VIF checks if the variable is not orthogonal to the other variables in the model. If VIF f 10, i.e., if 1/VIF p 0.10 then there is a problem with the model that needs to be corrected. The study adopts both measures, VIF and Tolerance, to investigate multicollinearity.

On the other hand, besides multicollinearity, the Goodness of Fit of the model needs to be ascertained. Preference is here given to the Hosmer-Lemeshow test since it does not depend on the number of trials per row in the data as the other goodness-of-fit tests (e.g., Deviance test, Pearson test) do. The test, based on the Chi-square test, is thus a more trustworthy indicator of how well the model fits the data when the data have few trials per row. Procedurally, the measure involves testing the null hypotheses Ho: “the model is a good fit”, against the alternative Ha: “the model is not a good fit”. A significance level <0.05 indicates that the null hypothesis is accepted, otherwise we reject the null and accept the alternative. In other words, if all of the goodness-of-fit tests have p-values higher than the usual significance level of 0.05, then the tests fail to provide evidence that the predicted probabilities deviate from the observed probabilities in a way that the binomial distribution does not predict.

It is evident from Table 5 that the education level is significantly important in youth unemployment in Uganda. Specifically, in the current case, the marginal effect at the means for primary of 0.154 tells us that the predicted probability of being unemployed is 0.154 greater for the individual with a primary level of education than for the one who has no education. Also, being a youth who attended secondary education increases the probability of being unemployed by 0.229 compared to youth with no classroom experience. It gets worse for higher education, where the marginal effect at the means for the post-secondary level of education indicates that the predicted probability of being unemployed is 0.411 concerning the youth with “no education.”

By implication, the higher the level of education the more likely one is unemployed. Perhaps this could be explained by several reasons including but not limited to poor quality education, inadequate job-matching skills, job preference, job availability, slow growth of the private sector compared to the number of graduating individuals per annum. Nevertheless, the finding conforms with Nganwa et al (2015) who found that the higher the level of education the more likely one is unemployed.

To further elucidate on the education factor, Table 6, focusing on marginal effects, reports that having attended school as a female in primary, secondary, or post-secondary increases the probability of being unemployed by 0.100, 0.160, and 0.338 respectively at 1% compared to those without education. On the other hand, being a male youth who attended primary, secondary or post-secondary increases more the chances of being unemployed by 0.188, 0.318, and 0.542, respectively. Intuitively, a female youth in the categories of those with certificates in primary, secondary, and post-secondary is more likely to get jobs compared to their male counterparts with similar qualifications. Clearly, the finding suggests that education level increases with the unemployment rate but has stronger adverse effects for the male youths relative to their female counterparts. A previous study by Nganwa et al. (2015) concurs with our documentation.

Also, important to observe in Table 6 is that if we have, say, three female youths with primary, secondary, and post-secondary levels of education, respectively, these are likely to witness their probabilities of unemployment increase by 10%, 16%, and 33.8%, respectively, compared to a youth with no education. Yet, on the other hand, three male youths with the education level of primary, secondary and post-secondary, respectively, are likely to experience much higher and significant increases in the probabilities of being unemployed by about 18.8%, 31.8%, and 54.2%, respectively, compared to male youth with no education. We take this to mean that male youth are more likely to face unemployment longer than their female counterparts with similar education levels. This may underline the fact that females are more likely to find employment in the informal economy, characterizing the Ugandan economy, than males, as earlier pointed out in ILO (2004).

Turning to the next specific objective, Table 5 suggests that being male (compared to being female) increases the probability of being unemployed by about 5.5%. The result of logistic regression shows that sex is an important determinant of youth unemployment in Uganda. The result reveals that being male increases the probability of getting unemployment by 0.055 compared to the female counterparts and it is statistically significant at 1%. The finding is in line with the previous authors Baah-Boateng (2012) who found unemployment to be a bigger labor market challenge for men than women and Isengard (2003) who found that young males in the United Kingdom are at a higher risk of becoming unemployed compared to their female counterparts because of the growth of service sector employment which has offered part-time or flexible jobs for female gender but it is inconsistent with the findings of Dickens and Lang (1995) who found higher unemployment rates among women than men.

The gender factor is explored further in Table 6 which provides evidence that females in households under the headship of a married female decrease the probability of being unemployed (compared to female youth living in a household with an unmarried female head) by 5.8%. This outcome is highly significant at 1% level. On the other hand, being male youth in a household headed by a married female doesn’t affect in any significant way one's unemployment status in comparison to being a male living in a household headed by a single female. However, a female youth, living in a household headed by a divorced female head, is likely to witness a reduction in the probability of being unemployed by 6.7% compared to the female youth staying with the unmarried female head of the household. The corresponding reduction in the probability of being unemployed for a similar female youth but living in a household headed by a male is 6.8%, in comparison to a female youth staying in a household headed by an unmarried female. On the other hand, a male youth in a household under the headship of a divorced female is likely to witness an increase in the probability of being unemployed by about 4.5% when compared to one under the household headed by an unmarried female. In the case of male youth staying in a household headed by a male, however, the probability of being unemployed increases by only 2%, under a similar comparison. While a similar male youth living in a household headed by a widow has his probability of being unemployed likely to increase by 2.7% compared to the male youth staying in a household headed by an unmarried female, there is no evidence that a female youth leaving under same circumstances will have her probability of being unemployed affected at any significant level.

It is interesting to note that a female youth who is married would likely have a reduced probability of unemployment by about 0.022, while a female youth who is divorced experiences an increase in her chances of being employed by 0.042 compared to an unmarried female youth, though the significance in levels of the latter outcome is weak (10%) whereas in the former case it is at 1% statistical level. We fail to find evidence for the male youth counterpart in similar marital status. On the other hand, Table 6 demonstrates that a male youth living in a household where the education level of the household head is primary is likely to witness an increase in the chances of being unemployed by 0.013 compared to a male youth living in the household headed by a person who has never received any formal education, though the outcome is weakly significant at 10% statistical level.

Further, gender differential effects can be observed in terms of regions. Table 6 exhibits results that reveal that being a female youth from Eastern, Northern, and Western parts of Uganda (compared to the Central) would reduce the probability of being unemployed by 2.9%, 2.7%, and 4.1%, respectively. On the other hand, being a male from the respective locations would reduce the probability of being unemployed by 2.1%, 2.5%, and 1.2%, respectively, taking the Central region as a reference point. In other words, females are more likely to be employed compared to their male counterparts in the Eastern Northern and Western regions. For the case of males in the Western region, however, data is not committal. Females having a greater probability of getting out of the unemployment net faster than males in these regions could perhaps be attributed to the female readiness to undertake any job available especially if they have been unemployed for long. It is also possible that the nature of the labor markets in these regions compared to the Central region would favor more females than males. For example, according to Bowen et al. (2015), a higher proportion of women than men work in farming in Uganda —76% versus 62%. A full explanation for this scenario is outside the scope of our study. For now, it suffices to note that findings in line with regional disparities are not unique. For example, Kingdon and Knight (2004) find out that in South Africa, Black workers in all provinces except northern and eastern Cape are significantly less likely to be unemployed than those in Gauteng.

Regarding the residence factor, Table 5 shows that if a youth is from an urban setting, that youth's predicted probability of being unemployed is 0.041 in comparison to a youth in the rural area. The outcome is highly significant at 1% statistical level. The result means that urban youth are likely to be unemployed than rural counterparts, after controlling for region, marital status, age, health, sex, and education of the household member as well as the education, marital status, and sex of the household head. Perhaps it is easier for the youth to be employed in rural areas than in urban areas in Uganda especially in the agriculture sector due to the informal nature of employment in rural areas. In urban areas, youth are more constrained in formal employment requirements such as education, skills, and experience which most youth do not possess. This finding is in line with previous authors (Msigwa and Kipesha, 2013) who found unemployment among the youth to be high in urban areas compared to rural areas, but they also corroborate the argument in Ahaibwe and Mbowa (2014) that urban youth are more likely to be unemployed (12%) than rural youth (3%).

In Table 6, the estimated marginal effects show that residence has a significant effect on youth unemployment in Uganda. The computed result shows that urban female youth increase their chances of unemployment by 0.024 compared to their rural counterparts while male ones increase the probability of being unemployed by 0.021 and this is statistically significant at 1%. The difference between females and males in urban areas is that urban male youth at 1% are likely to be employed than their female counterparts. The reason is likely about the nature of jobs in urban, job preference and availability and, gender disparity which favor the male youth than female ones. This is consistent with the finding of Dickens and Lang (1995) who found the unemployment rate higher in women than men. The result disagrees with the finding of Mulu (2012) who found that a bigger percentage of unemployed women (70.74%) resided in rural areas.

The estimated results show that age has a significant effect on youth unemployment in Uganda. Specifically, in Table 5, the relevant marginal effect of 0.005 on age, which is highly significant at the 1% level, demonstrates that age is positively associated with youth unemployment. Intuitively, as the age of the youth increases, the more one is likely to stay in the unemployment net. The result is consistent with the finding by Biagi & Lucifora (2008) who argues that changes in the population age structure are positively related to the unemployment rate of young workers, but have no effect on adults. Perhaps, one possible explanation would be that as years increase without a job, the unemployed youth has less skills and is even likely to fail job interviews due to the possibility of forgetting what he/she learnt earlier and therefore more likely to remain in the unemployment net. It can also be argued that for the youth who continue in school and therefore add extra years to their age, these become more educated and are often biased toward wage-paying formal jobs that are less supplied in the labor market. This is in line with our earlier finding that as one upgrades from primary to secondary to post-secondary, the possibility of remaining unemployed gets higher. By implication, as one advances in age, the probability of remaining unemployed gets bigger. The positive association between age and unemployment is likely to be the consequence of a “disillusioned worker” effect as it can be argued that the longer an individual is economically inactive, the harder it is to re-enter employment (Vitanen, 2001). For, as the latter author contends, the human capital deteriorates with duration and number of spells of unemployment.

For clarification of the finding in Table 5 on age, Table 6 shows that an increase in the age of a female youth is likely to significantly stimulate the probability of being unemployed by only 0.3% at 1% statistical level. Whereas the effect is significant for females, there is no evidence for the males. By implication, as a female's age increases the more likely she will drop out of the unemployment net. Perhaps by then, the female is taken over by family responsibilities with less time for paid work, especially given the strong cultural beliefs putting the woman in the kitchen. The finding is however not new as Carlsson and Eriksson (2017) similarly record that older workers and women often have lower employment rates, and the duration of unemployment increases with age.

Tables 7–10 exhibit results regarding the interaction effects of education, gender, residence, and age respectively, each with the rest of the variables. Several interesting findings deserve attention. For example, in Table 7, the role of the education level of the youth is found to be significantly dependent on marital status and region. Specifically, a primary level educated youth, in comparison to an uneducated one, will have the probability of being unemployed reduce if the youth is married. For a secondary school level youth, however, the probability of being unemployed reduces if the youth is divorced compared to one with no education. The possible intuition is that one struggles to accept any job to survive since one no longer depends on the divorced partner. If a youth falls under the category of post-secondary level and is a widow or married, the probability of being unemployed falls in either case in comparison to one without education. Similarly, being from the Northern region is associated with positive effects of primary education on youth unemployment. This moderating impact is significant at a 5% statistical level.

On the other hand, data shows that in comparison to the youth with no education, the youth with post-secondary education are likely to experience an increase in their unemployment status if they are from the Eastern or Northern regions, though the effect appears much larger for youth in the northern region than for those from the East, albeit weakly significant either way. Perhaps the labor markets in these regions do not attract higher-level skills attained by the youth in the post-secondary category. One could rightly argue that since most of the youth with post-secondary education appear to prefer white-collar jobs, which are not readily available in the rural regions such as the East and Northern regions, most of the youth with such education levels are likely to end up remaining unemployed. By implication, there is little demand for labor with education levels beyond secondary school education in the East and Northern region. It is thus not unimaginable that the differences in the unemployment rates across regions could be because different regions have very different labor markets and some have fewer jobs available. This corroborates what is recorded in Merotto (2019) where the author notes that the density of paid (wage) employment increased between 2002 and 2014, but much more in the Kampala, western, and central regions than in the eastern and northern regions.

In Table 8, we provide evidence that the effect of gender on youth unemployment is also indirect via education, marital status, and region. For example, it is observed that compared to the female youth, a male youth in the secondary education category is likely to experience an increase in the probability of being unemployed relative to one with no education. Similarly, a male youth who is married is more likely to witness an increase in the probability of being unemployed compared to an unmarried female. But also, being a male from the Western region increases the likelihood of being unemployed compared to a female youth from the Central region. Furthermore, being a youth from the urban area reduces the probability of being unemployed if one is a male, compared to a youth from a rural residence. Similarly, a youth from the urban residence but with a post-secondary level of education is likely to have a reduced likelihood of being unemployed, just as an urban youth that is married or divorced. However, if the youth is from an urban area in western Uganda, the likelihood of being unemployed increases compared to the rural youth.

From Table 5, we take note of the effects of the demographic factors, viz., marital status and region of the youth as well as the age, education, marital status of the household head, that we included as control variables in our models. For example, while marital status is observed to significantly determine youth unemployment of young females it does not affect male youth in Uganda. The estimated result shows that being a married female youth reduces the probability of being unemployed by 0.034, compared to the unmarried youth, and where the marginal effect at means is highly significant at 1% statistical level. The estimate also shows that in comparison to an unmarried youth, being a divorced female youth increases the probability of being unemployed by 0.051 where the observed coefficient is significant at 5% conventional levels. By implication, married females have more employment opportunities than the divorced ones. This is in agreement with the finding of Verhaeghe et al. (2012) and Zhang ad Zhao (2011) who found that married women have a higher LFPR than single women do. Axelrad et al (2018) similarly document that divorced and widowed men are less likely than single men to find a job.

Still notable in Table 5 is the observation that being a youth living under a household headed by a married female significantly reduces the probability of being unemployed compared to the youth living under the household headed by an unmarried female. The associated marginal effect at the means is −0.059 and significant at a 5% statistical level. This is in line with the argument of Neethi (2017) that on average, the married-couple households normally tend to be well-off and with connections or social networks that can assist the unemployed youth get out of the unemployment gap compared to a single female-headed household. On the other hand, an unemployed youth living in a household headed by a male is likely to experience an increase in the probability of being unemployed by 0.063 compared to a youth under an abode headed by an unmarried female. This outcome is significant at a 5% conventional level. Perhaps, this is because the unemployed youth in the household feel more secure under a male head that could be providing the requirements at home and can afford not to work while they search for a job. The evidence at hand that a female youth living under a household headed by a divorced female or under a male-headed household is likely to have her chances of being unemployed reduced, whereas the opposite appears to be the case for the male youth living in either household, could be an indication that perhaps females as much more favored by either household heads, helping them to get employed faster than males. The explanation for the outcome could however lie elsewhere, and therefore a subject of further study. Nevertheless, relatedly, Moepeng and Tisdell (2008) has previously argued that unemployment is much higher for female heads of household than for male heads.