Decent Rural Employment in a specialised and a diversified production system in Tanzania

ABSTRACT

The agricultural sector in developing countries plays a vital role in food security as well as providing employment opportunities to the rural population. This study examines how decent quality of rural employment can be associated with technical efficiency of agricultural production of smallholder farmers in Tanzania. While labour is considered an input in the production process, various forms of labour are rarely measured in the context of employment quality. Using a latent-class stochastic frontier model, two types of farming systems are identified: a specialised crop system and a diversified farming system. The study found child labour to be significantly contributing to the inefficiency of agricultural production only in the diversified farming system, while precarious employment contributed to the inefficiency in both farming systems. Based on these findings policymaking that targets decent employment in developing countries needs to account for differences in farming systems.

KEYWORDS:

• Sub-Saharan Africa
• efficiency analysis
• quality of employment
• smallholder farmers
• child labour
• farming system

1. Introduction

Productive employment, either waged or self-employed, is necessary for improving rural livelihoods as the earnings from labour are often the main source of income for poor rural communities and hence are contributing to poverty reduction (Ayenew et al., 2017). Farming in Sub-Saharan Africa (SSA) is predominantly of small scale. Around 80% of farms in SSA own less than 2 hectare of land (Wiggins, 2009; Lowder et al., 2016) with the majority of them being used for integrated rainfed mixed crop and livestock systems (Thornton & Herrero, 2015). Acknowledging the diversity in production system-specific agricultural performance is important as each farming system allocates its resources differently. This is even more so in Africa with its variety of differences in farming systems, as all types of production systems have different labour requirements and employment opportunities (Seo, 2010; Garrity et al., 2012; Gebrekidan et al., 2020).

The decency of employment is an important factor in determining the quality of livelihood. In particular, the FAO underlines the relevance of Decent Rural Employment (DRE) as crucial for poverty reduction strategies in rural areas (FAO, 2010, 2012). Ayenew et al. (2017) investigated the impact of DRE on technical efficiency of agricultural production by testing several indicators developed by Ghai (2003), to proxy the core dimensions of DRE, such as precarious employment and child labour. Technical efficiency of agricultural production measures the effectiveness with which agricultural inputs are converted into agricultural outputs and consequently constitutes a measure of agricultural economic performance.

In the Ayenew et al. (2017) paper some of the indicators examined, like the employment-to-workforce ratio, did not show a statistically significant influence on agricultural technical efficiency. We expect to find different results when looking into specialised vs. diversified farming systems, especially when their production functions are estimated individually, because it is commonly assumed that similar farms share similar production functions but each production system is characterised by different labour dimensions (Otte & Chilonda, 2002). Therefore, the present study categorises farms into similar farming systems to explore whether DRE indicators can affect household-level performance among the different farming systems. This study is probably the first to establish an empirical relationship between DRE indicators, technical efficiency of agricultural production and differences in the farming system. This paper was written in collaboration between the Technical University of Munich (TUM) and the Food and Agricultural Organization of the United Nations (FAO). This work is based on a technical report from this collaboration (Schickramm et al., 2016).

To classify between different farming systems in Tanzania, the study considers farm diversification, the number of livestock used and the environmental characteristics in the context of agricultural production. First, the production systems and technologies for a sample of farms are identified using a latent-class method. The farming system refers to the mix of activities conducted on the farm, while the production technology refers to how the different inputs are transformed into outputs. This methodological framework supports the estimation of technical efficiency for diverse farming systems and also explores how DRE indicators can impact the technical efficiency of different farming systems. The data used in this study are derived from the 2010–11 report of the Living Standards Measurement Study (LSMS) for Tanzania from the World Bank.

To explore this further, Section 2 provides a brief overview of the concepts of DRE and farming systems, and describes the correlation between these two concepts. Section 3 introduces the empirical model and the estimation methodology used. Section 4 presents the different latent classes and the results of the estimation. Section 5 summarises and discusses the results by highlighting the main implications concerning DRE for the existing policy debates on the reduction of rural poverty and the future role of farming systems in SSA.

2. Decent rural employment, farming systems and the conceptual framework

A large proportion of the impoverished rural population depends on their own labour for their livelihoods and spends a substantial part of their daily time working. Therefore, the International Labour Organisation (ILO) proposes that providing better and decent work opportunities is as important as generating productive and gainful employment opportunities. It defines Decent Work as ‘opportunities for women and men to obtain decent and productive work in conditions of freedom, equity, security and human dignity’ (International Labour Organization, 1999). The FAO applies the DRE concept to the rural context, where employment is diverse and it refers to any activity performed for pay or profit, including paid, self-employed and family labour (FAO, 2012). Promoting DRE, the FAO operates on four main objectives of the so-called Decent Work Agenda, with gender equality as a common objective. These four objectives are (1) employment creation and enterprise development, (2) working conditions and social protection, (3) rights at work and (4) employees’ and employers’ organisation and social dialogue.

The first objective ‘employment creation and enterprise development’ focuses, as the name suggests, on the economic activities that surround employment in rural areas. This can range from educational programmes regarding better agricultural performances to promoting agribusinesses, as well as any other employment enhancing activity. The second objective ‘working conditions and social protection’ refers to anything related to the working environment. This includes wages, occupational health and safety measures (OSH), as well as the provision of labour saving technologies and care services. These can be especially important for households that are burdened by diseases or other care tasks as these can significantly reduce the workload of women in those households. Maternity protection and minimum wage in agriculture are also part of this objective. The third objective ‘rights at work’ deals with discrimination and rights in the working environment. This includes the abolishment of child, as well as forced, labour. Additionally, it deals with the ending of any form of discrimination, such as age, race, gender or religion. In rural settings, the fourth objective ‘governance and social dialogue’ strengthens rural workers and producers in their ability to take action and participate in the decision-making process. Especially women and youth are encouraged to be actively engaged in decision-making as they are considered vulnerable groups (FAO, 2010).

The diversity and complexity of these four objectives of the Decent Work Agenda complicates the attempts to assess decent employment, as for the same objective various indicators and measurements may apply. Data availability has determined the indicators used in the literature and this study. However, the definitions used in data collection vary spatially between countries and dynamically within a country over time. To avoid these spatial and temporal inconsistencies, the present study focuses on one country at one time point only. The indicators used in this study for different components of decent work should still be interpreted cautiously providing an approximate measure of performance (Anker et al., 2003; Ghai, 2003).

The present study adapts a set of indicators used in the study of Ayenew et al. (2017) and Ghai (2003) by examining the relationship across different latent classes of farming systems. The definition of farming systems has changed greatly over the past 50 years. Earlier, scholars limited their emphasis to the farm alone, the crops produced and the household’s food security status. In the long term, the focus shifted towards a more holistic approach that included household-, community- and district-level characteristics as well as different forms of livelihood strategies. The relevance of gender- and efficiency-related, as well as sustainability and environmental issues has also increased over time (Dixon et al., 2001; Therond et al., 2017). Existing farming systems are characterised by individual resource endowment, family or household circumstances and resource flow and interaction among the biophysical, socio-economic and human factors (Dixon et al., 2001). As this set of characteristics varies greatly from farm to farm, the present study adopts the definition by Dixon et al. (2001) to classify various farming systems. Dixon’s definition considers a farming system as ‘a population of individual farm systems that have broadly similar resource bases, enterprise patterns, household livelihoods and constraints, and for which similar development strategies and interventions would be appropriate’ (9).

The increasing population also adds to the pressure on farming systems in addition to decreasing size of smallholder farms over time (Jayne et al., 2014). Rural, as well as urban population, has been growing fast in Africa leading to increasing food demand, which subsequently entails the intensification of agricultural production to meet these needs (Giller et al., 2011). Intensified agricultural production usually requires a change in farming performance towards high efficient production systems. Scarcity of land calls for the intensification of land use in all farming systems; however, considering each farming system separately can give a better insight of the agricultural sector for policymaking. There are various classifications of agricultural household depending on the focus of the study. We are going to establish production systems based on a latent-class approach. The latent classes are built around livestock ownership, diversity of production and an environmental characteristic.

Measuring technical efficiency of different types of farms is not a novel approach in European studies. For example, in organic versus conventional farming by Tzouvelekas et al. (2001) in Greece and Bremmer et al. (2002) in Finland. Sauer & Paul (2013) adopted a latent-class approach to classify a sample of Danish farms. These examples consider labour as an input measure when considering productivity; however, they do not classify different forms of labour or consider the quality aspects of employment.

While the majority of studies assess decent employment at the country level, the present study pays special attention to the household-level decent employment adjusting the indicators proposed by Ghai (2003), such as Ayenew et al. (2017), to the household level. Table 1 summarises these indicators and their measured effects, as presented by Ayenew et al. (2017) and provides the hypotheses and ideas for the current study.

Table 1. Decent Rural Employment indicators and effect on efficiency.

Objective	Indicators used	Outcomes in Ayenew et al. (2017)	Crop-based production system	Livestock-based production system	Diversified production system
	measurement
Objective 1: Employment creation	Employment-to-total workforce ratio	Insignificant for Tanzania	Positive/negative	Positive/negative	Positive/negative
	Proportion of employed HH members to total HH workforce available
Objective 2:Social protection	Share of government transfer to income	Positive	Positive/negative stronger	Positive/negative stronger	Positive/negative weaker
	Total transfer from government and NGOs in Tanzanian Shilling from the total income
Objective 3: Standards and rights at work	Child labour ratio	Negative	Positive/negative	Positive/negative	Positive/negative
	Proportion of child labour from the total labour used for agricultural activities by the HH
	Precarious employment ratio	Negative	Negative	Negative	Negative
	Proportion of HH seasonal and casual labour from the total HH agricultural workforce

Under objective 1, the ratio of employed household members to total household workforce¹ is used to measure the availability of employment opportunities. Ayenew et al. (2017) found no major impact of employed workforce on farm productivity in the case of Tanzania. When analysing the technical efficiency of agricultural production of different farming systems, it is likely that the effects of DRE indicators on technical efficiency might differ depending on farm characteristics. The effect of the employment outside of the agricultural household usually negatively affects on-farm income. However, Taylor et al. (2003) found no negative effect on yields. Earnings from off-farm employment additionally compensated for the effect of the labour loss. If Tanzania was to have more labour than it required for their on-farm activities, there are only positive effects on technical efficiency of agricultural production from working outside of the household (Ruttan, 2002).

For objective 2, on social protection, indicators capturing access to cash and food transfers are applied. Smallholder producers and rural population, especially considering the limited outreach of insurance markets in the rural areas of SSA, benefit considerably from such programmes. In addition, informal and governmental transfers can positively affect technical efficiency of agricultural production. More specialised farming systems may be greatly affected by these payments, as they can be liable to a complete loss of earnings due to climatic extremes and/or market developments. However, in a diversified production strategy, earnings are spread over several activities and an income loss in one activity would have a less severe effect on the total household income (Benjamin & Sauer, 2018). There is also a significant effect of knowledge transfer and learning from government programmes.

Objective 3 is proxied by two indicators that capture forms of employment, which are non-desirable or ‘non-decent’, namely child labour and precarious employment. Prevalence of child labour and precarious employment in agriculture presumably negatively affect the technical efficiency of agricultural production. Both are measured as proportions with respect to the total workforce engaged in agricultural production. In Tanzania, it is common for children to participate in agricultural activities at the household level. In this study, child farm activity is measured as the proportion of child labour to the overall agricultural labour force, indicating the extent of reliance of a household on child labour for the agricultural production. According to Ayenew et al. (2017), both child labour ratio and precarious employment ratio had statistically significant negative effect on technical efficiency. The effects of child labour might differ across farming systems because it is predominantly employed for herding activities (FAO, 2013). Across all farming systems poor families could gain from child labour in the short run as child labour can be a cheap replacement for adult labour (The World Bank, 2007). In the long run, productivity loss due to foregone learning is undebatable. In Malawi, for example, using adult labour for herding activities is perceived to be inefficient use of labour potential (FAO, 2013). However, the use of child labour can only be more efficient if they are able to replace the work of a grown woman or man. This will depend on the type of crop or livestock work, as children could be substituting for adult labour in those activities that are perceived to require less skills or physical ability. Hence, in all the three farming systems, a positive effect in the short run or a negative effect in the long run of child labour on productivity is perceived, depending on whether child labour can be seen as a good substitute for adult labour. It is, however, expected that children can substitute adult labour in herding activities easier than in crop-related work.

As family members in rural areas of SSA handle the majority of farm activities, there are only limited seasonal and casual jobs left for the rural landless and other resource poor workers. These low-paid, precariously employed workers have few incentives and low motivation to improve efficiency of agricultural production. Precarious employment is expected to negatively affect all three production systems, as the problem of moral hazard arises (Bhalotra & Heady, 2003).

For the fourth objective of DRE, related to employees’ and employers’ organisation and social dialogue, no indicators are available yet in the LSM Study; therefore, the fourth objective has been excluded from the present empirical study. Indicators from other studies cannot be used as it would be impossible to link them to the correct households used in this study.

3. Methodology

Structural changes and technical patterns related to the agricultural activities can only be correctly assessed and interpreted when farms operate under different technologies. The production technologies of farms, or generally agricultural decision-making units, can be characterised by exogenous factors, such as the production method (Kumbhakar et al., 2009). Classifying farms is relatively novel; for example, Chambers et al. (2011) classified farms based on the amount of rainfall. Moreover, several empirical studies have classified farms based on organic and conventional agricultural typologies (Tzouvelekas et al., 2001; Lansink et al., 2002; Latruffe & Nauges, 2014). This approach focuses on observable characteristics, which sometimes may ignore unobservable factors and consequently incompletely display the whole picture. Instead, a clustering procedure that considers exogenous unobserved characteristics can be used to shed more light on the set of factors – both observable and unobservable – that can affect the agricultural production decision-making process. In particular, a multivariate latent-class model can be applied, which is based on unobserved characteristics (Magidson & Vermunt, 2001; Sauer & Paul, 2013) and can be linked to different regression-type modelling procedures.

The present study uses a two-step approach: first, a production frontier for the agricultural production of the farm is estimated using a latent-class approach giving three variables to distinguish between the different production technologies. Tropical livestock unit (TLU), concentration index and annual precipitation are the three indicators used leading to the development of two distinctive latent classes. The first latent class (T1) presents a specialised crop farming system, as farms in this latent class are not involved much in livestock activities and have less land and labour on average compared to the other latent class. Farms in latent class two (T2) show a higher livestock index and a lower concentration index suggesting that farm households in T2 are less specialised than in T1. In addition, in this farming system, higher livestock index is combined with higher land and labour resource availability. Therefore, the T2 system can be viewed as a diversified farming system. As we do not have a specialised livestock farming system in this study, we will refer to the specialised crop farming system as the specialised system (T1) and T2 will be referred to as the diversified system. In the second step, the posterior probability to belong to one of those latent classes is used to estimate two distinct production frontiers with explanatory variables to explain the technical efficiency of agricultural production for each latent class.

As multiple outputs are produced by the farms in the study sample, and considering the effect of DRE on the technical efficiency of agricultural production based on the outputs that is produced, a distance function production frontier approach is appropriate (Orea & Kumbhakar, 2004; Coelli et al., 2005; Newman & Matthews, 2006; Rahman, 2009).

In the distance function, the technical efficiency refers to the distance between the farms’ production and the optimal possible production given the inputs that the farm is using.

The multi-output distance function framework enables the researcher to isolate the effect of DRE on the farm-level technical efficiency for each of the agricultural outputs produced (crop and livestock production). As the technical efficiency is the distance to the optimal, it is actually an inefficiency, which we are trying to explain using different factors that can play a role in determining how (in)efficient a farm is operating. Those different factors that can be used to explain (in)efficiency across different farms are regional dummies (used as an explanatory variable to capture unobservable characteristics with respect to space), weather characteristics, age and sex of the household head, age dependency ratio,² livestock holding in TLU, access to extension services, concentration index of the agricultural production, access to credit, distance to the nearest road, and the set of DRE indicators (as defined in Section 2, Table 1). Table 3 shows the detailed summary statistics for the diversified (T2) and specialised farming system (T1).

The data comes from the Living Standard Measurement Survey – Integrated Survey on Agriculture (LSMS-ISA) 2010–11. It was collected by the Tanzanian National Bureau of Statistics (NBS). The sample is according to the World Bank Group representative for the nation as a whole, and provides reliable estimates of key socio-economic and agricultural variables. The total size for this sample is 3924, after excluding urban, missing and unreliable observations a total of 931 household remains.

This framework offers policy makers the information to design and propose output-specific policy incentives for farmers. Please see the technical appendix for more details on the conceptual framework and specification of a multi-output distance function.

4. Results

In the specialised group of farms (T1), we have 303 agricultural households, while the diversified group of farms (T2) includes 628 farms (Table 2). We classified the first group of farms (T1) as specialised, as they are involved in activities requiring very little livestock, less land and less labour, on average, compared to the diversified group of farms (T2). Consequently, the first group of farms (T1) is classified as a specialised crop-based farming system with a low livestock capacity.

Table 2. Summary statistics of the latent classes and the overall sample.

Latent Class	Variable	Mean	Std. Dev	Min	Max	Obs.
T1 specialised farming system	Tropical livestock index	0.151	0.597	0.000	570.000	303
	Land	2.842	3.590	0.0405	284.292	303
	Labour	155.425	141.278	240.000	887.000	303
	Concentration index	1.483	0.468	0.098	199.774	303
	Precarious Employment ratio	0.074	0.159	0.000	0.872	303
	Employment-to-workforce ratio	0.791	0.290	0.000	1.000	303
	Child labour ratio	0.065	0.127	0.000	0.612	303
T2 diversified farming system	Tropical livestock index	2.672	7.866	0.000	118.800	628
	Land	3.607	5.811	0.0445	655.307	628
	Labour	170.342	163.523	200.000	1236.40	628
	Concentration index	0.860	0.480	0.032	199.615	628
	Precarious Employment ratio	0.101	0.180	0.000	0.985	628
	Employment-to-workforce ratio	0.817	0.251	0.000	1.000	628
	Child labour ratio	0.057	0.115	0.000	0.668	628
Whole sample	Tropical livestock index	1.840	6.561	0.000	118.800	931
	Land	3.338	5.199	0.040	655.307	931
	Labour	164.360	156.718	200.000	1236.40	931
	Concentration index	1.064	0.558	0.003	199.774	931
	Precarious employment ratio	0.093	0.174	0.000	0.985	931
	Employment-to-workforce ratio	0.808	2.265	0.000	1.000	931
	Child labour ratio	0.0596	0.119	0.000	0.668	931

In the second group of farms (T2), farm households show a higher livestock index combined with a lower concentration index in comparison to the first group of farms (T1). Additionally, more land and labour, on average, are available. This suggests that farm households in T2 are less specialised than in T1 and have more livestock activities, higher land and resource availability. Therefore, the system can be described as a diversified farming system (T2). When we look at the labour patterns in the sample, we find that the first group of farms (specialised farming system) relies less on precarious employment and more on child labour compared to the diversified farming system (T2). Both share a similar employment-to-workforce ratio.

Table 3 presents the maximum likelihood estimates of the multi-output distance production function for both of the groups of farms. The variables land, labour, intermediate inputs and livestock crop ratio are statistically significant from zero and show the expected signs. LIMDEP software has been used for the maximum likelihood estimations.

Table 3. Maximum likelihood estimates for the two latent classes.

	Latent Class T1 Specialised farming system		Latent Class T2 Diversified farming system
Production Variable	Coefficient	Standard Error	Coefficient	Standard Error
Constant	0.880***	0.077	0.317***	0.087
Land	−0.160***	0.032	−0.238***	0.021
Labour	−0.394***	0.035	−0.253***	0.028
Intermediate Inputs	−0.121***	0.019	−0.199***	0.016
Livestock crop ratio^a	0.309***	0.013	0.130***	0.011
Land^2	0.042**	0.021	−0.014	0.013
Labour^2	0.001	0.026	−0.014	0.021
Intermediate Inputs²	−0.032***	0.008	−0.018**	0.007
Land x Labour	0.019	0.035	−0.005	0.026
Land x Intermediate inputs	−0.060***	0.019	−0.0004**	0.014
Land x Labour	0.022	0.020	−0.005**	0.017
Land x Livestock crop ratio	0.131***	0.016	−0.022	0.010
Labour x Livestock crop ratio	−0.066***	0.014	−0.010*	0.013
Intermediate inputs x Livestock crop ratio	−0.028***	0.008	0.011	0.007
Inefficiency Determinants
Age of the household head	−0.243	0.200	0.003	0.003
Employment-to-workforce ratio	−6.301	6.189	0.124	0.154
Precarious employment ratio	21.234*	11.479	1.270***	0.305
Precipitation of the wettest quarter	0.007	0.007	0.001	0.0003
Age dependency ratio	−4.215	3.453	−0.049	0.064
Women to total labour ratio	1.048	6.587	0.162	0.239
Distance to the major road	0.039	0.050	0.004**	0.002
Sex of the household head	−7.666	7.023	−0.228	0.153
Household head literacy	−5.685	5.348	−0.388***	0.148
Region	−0.609*	0.366	−0.019**	0.009
Advisory service	−2.072	9.723	0.022	0.226
Share of government transfer to income	−194.290	379.775	−37.893	21.014
Access to credit	−12.162	21.616	−0.175	0.378
Child labourer to total labour ratio	27.518	32.129	0.812**	0.413
Other Parameters
Lambda	4.987	3.442	163.131	0.232
Sigma	3.293	2.177	0.816	0.043

^a Amount of income gained from livestock divided by the amount of income gained from crop production.

In the specialised farming system (T1), precarious employment has a significant negative³ effect on the technical efficiency of the agricultural production process. This implies that a higher precarious employment ratio on the farm correlates with a lower technical efficiency. A significant effect on the inefficiency of farms in the diversified farming system T2, can be observed for: precarious employment, literacy of the household head, distance to the road and child labour ratio. Higher ratios of precarious employment and child labour are associated with lower efficiency levels for agricultural production. If the household head is literate, this significantly contributes to higher farm-level efficiency. When we look more closely at precarious employment, which is significant for both systems, we find that the magnitude of the coefficient is 20 times higher in the specialised farming system (T1) than in the diversified farming system (T2). A higher seasonality of production in the specialised system due to the crop production could be the reason for this. Also, the region where the farms are located can have a significant effect on both farming systems.

5. Discussions and conclusion

Research of the relationship between DRE indicators in the context of technical efficiency has been scarce. The present study appears to be the first to address this gap by establishing an empirical relationship between indicators that act as proxies for dimensions of DRE and technical efficiency for different farming systems. This study provides the opportunity to compare two different farming systems. We observed both a specialised (crop) production system and a more diversified farming system. In the study by Thornton & Herrero (2015), the diversified farming system also represents the majority of the sample. Due to the data availability, the diversification within the farming system refers exclusively to on-farm diversification, leaving enough scope for future research on a complementary investigation of the implications in terms of off-farm diversification.

Regarding the effect of the DRE indicators considered in the present study, we found no significant effect of the indicators used for objective 1, employment creation, and objective 2, social protection. Precarious employment, which is part of objective three ‘rights at work’, always led to lower technical efficiency of agricultural production no matter the farming system. The effect, however, is a lot greater for the specialised production system, with the coefficient being almost twenty times the size than the one for the diversified production system. Precarious on-farm employment, which is characterised by the casualty of employment usually in combination with a low skill set of the labourer and low wages, can be associated with low production efficiency (Bhalotra & Heady, 2003; Ayenew et al., 2017). An interesting addition of this study is also that this form of employment may especially burden specialised crop producing farming systems.

Household head literacy in this study has a significant effect on technical efficiency of agricultural production in the diversified production system, but not in the specialised production system. One would expect that a higher literacy level allows farm household heads to make better production and marketing decisions. While we did not find a significant effect in the specialised production system, we believe that the role of skill development and education cannot be overemphasised in all production systems of developing countries, although it appears to be more critical in the diversified farming system.

When it comes to child labour, we find a significant effect on technical efficiency of the agricultural production in the diversified farming system, but not in the specialised farming system, as with literacy. This means that households that use a high percentage of child labour, in the diversified farming system, are less efficient than households that do not use child labour. In the diversified production systems, we see more livestock-related activities than in the specialised farming system. These livestock-related activities and limitations in family labour availability could have led to child labour in herding and animal care activities. This study shows that in addition to the negative long-term effects of child labour, there are also negative short term effects on the technical efficiency of agricultural production. While child labour could be a substitute for adult labour in some farm activities like herding, in this study child labour contributes to an inefficiency of agricultural production compared to adult labour. While the child labour percentage was higher in the specialised farming system, we could not prove a relationship with the efficiency of agricultural production of those farms, neither negative nor positive. From a policy perspective, this means that for the issue of child labour special attention needs to be given to specialised farms, as the percentage of child labour is greater without a proof that child labour is contributing to the inefficiency of agricultural production, making it even more necessary for policy attention.

It is important to consider here that based on the production function methodology, inputs (land, labour, capital) are converted into outputs (from livestock and crop) without taking into account the different costs of labour, which in the case of child labour is something to be considered as it is often freely available for the household. And this can be considered as a clear limitation of this study.

When it comes to precarious employment, it seems to be a challenge for both types of farming systems, although the effect is almost 20 times higher in the specialised farming system compared to the diversified farming system. Precarious employment needs appropriate policy attention not only for the technical efficiency of the farming systems that employ those labourers, but also for the labourers themselves that suffer from precarious employment conditions. Further empirical research on various farming systems may provide better information for policy makers in these countries on various DRE issues. It appears that DRE deficits in the agricultural sector of developing countries could be better addressed if these countries would implement production farming-oriented employment policies and strategies.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We have built this indicator adapting the ‘employment-to-population’ ratio to our analytical setting and data at disposal. Hence, employment-to-total workforce ratio is measured using the last 7 days as the reference period, includes those who were employed over the last 7 days reference period as self-employed, part-time, casual or seasonal work on farm/off/ or non-farm, after controlling for those who are inactive (too young and too old, went for schooling, ill and physically incapable).

2 The ratio of children and elderly on the normal aged population of the household.

3 In the table, the value is positive, as it indicates the relationship with the inefficiency. There is a positive influence of precarious employment on inefficiency, hence there is a negative impact of precarious employment on efficiency.

Technical appendix

A distance function can be represented in mathematical terms as: (1) $d_i^I=d^I\left(x_{1i},x_{2i}\dots x_{Ni,}{y}_{1i,}{y}_{2i}\dots x_{Mi}\right)$(1) (2) $d_i^o=d^o\left(x_{1i},x_{2i}\dots x_{Ni,}{y}_{1i,}{y}_{2i}\dots x_{Mi}\right)$(2) where Equations (1) and (2) illustrate the respective representation of an input- and output-oriented distance function (d_i) in a technological set of producing M number of outputs (y) using N number of inputs (x).

According to Kumbhakar et al. (2007), a production technology based on a distance function representation can be defined as: (3) $1=f\left(y,x,\text{β}\right)\exp \left(v+u\right)$(3) or in logarithmic expression (4) $0=\ln f\left(y,x,\text{β}\right)+v+u$(4) where y is the observed outcome, $ln\mathrm{f}\left(\mathrm{y},\mathrm{x},\text{β}\right)$ is the production frontier value pursued by the individual farm, and $v\sim N\left[0,{\sigma }_v^2\right]$ is the stochastic error term capturing the noise in the model. The inefficiency, the amount by which the farm fails to reach the optimum (the frontier) is represented by$u\sim N\left[{\alpha }^{\mathrm{\prime }}z,{\sigma }_u^2\right]$, with $u=\left|U\right|$, following a normal–truncated normal distribution which relaxes the restriction in the normal-half normal model of a zero mean (Greene, 2012). In this case, $u$ follows a distribution with mean, ${\mu }_u={\alpha }^{\mathrm{\prime }}z$, and variance$,{\sigma }_u^2$.

Battese & Coelli (1995) developed a single-step maximum likelihood procedure to estimate both the parameters of the distance frontier and the parameters of factors that potentially determine the technical efficiency of farms. Accordingly, this can be done by integrating the following equation of the mean of the one-sided error term ($u$) that captures the technical inefficiency into the estimation procedure. (5) ${\mu }_i={\alpha }_i{Z}_{ni}+{ϵ}_i$(5) where µ_i is the conditional mean of u_i from the first estimation procedure, Z_i’s are vectors of household parameters to explain the inefficiency parameter, ${ϵ}_i$ is the statistical noise, and α´s are the unknowns that will be estimated.

In contrast to the standard stochastic frontier approach, which fits one frontier for the whole sample, the latent-class stochastic frontier approach estimates a unique frontier for each latent class. In this approach, the basis of assigning a farm to a latent class is the highest probability (Orea & Kumbhakar, 2004). The posterior class probability is defined as follows: (6) $P_j\left({\delta }_j\right)=\frac{exp\left({\delta }_j^{\mathrm{\prime }}{q}_i\right)}{{\sum }_{J=1}^{J}exp\left({\delta }_j^{\mathrm{\prime }}{q}_i\right)},j=1,\dots .,J,{\delta }_j=0$(6) with the class probabilities being parameterised as a multinomial logit model, where $q_i$ is a vector of farm-specific variables. Based on the posterior probabilities from the latent-class function, we define the conditional likelihood function ($Lf\left(\theta ,\delta \right)$) as follows: (7) $Lf\left(\theta ,\delta \right)=\sum _{J=1}^{J}L{F}_j\left({\theta }_j\right)\ast P_j\left({\delta }_j\right),0\le P_j\le 1,\sum _j{P}_j=1$(7)

This approach allows a farm to use a combination of technologies, while the inefficiency of the farm can be weighted by the relative performance of the farm measured against all possible technological frontiers. Hence, having two reference technologies, we take into account the technology from every class. The number of classes is determined by the AIC and the BIC choosing the model with the lowest value.

Equation (8) describes Tanzania’s multi-output production function in a multi-output distance specification: (8) $\begin{array}{rl}-\ln crp\mid j=& ({\text{β}}_0+{\text{β}}_1\ln (liv/crp)+{\text{β}}_2\mathrm{l}\mathrm{n}(lan)+{\text{β}}_3\mathrm{l}\mathrm{n}(int)+{\text{β}}_4\mathrm{l}\mathrm{n}(lab)+{\text{β}}_5\mathrm{l}\mathrm{n}(tlu)+0.5{\alpha }_1(\mathrm{l}\mathrm{n}\left(lnd\right){)}^2\\ & +0.5{\alpha }_2\left(\mathrm{l}\mathrm{n}\right(int){)}^2+0.5{\alpha }_3(\mathrm{l}\mathrm{n}\left(lab\right){)}^2+0.5{\alpha }_4\left(\mathrm{l}\mathrm{n}\right(tlu){)}^2+{\alpha }_5\mathrm{l}\mathrm{n}(lan)\ast \mathrm{l}\mathrm{n}(int)\\ & +{\alpha }_6\mathrm{l}\mathrm{n}\left(lan\right)\ast \mathrm{l}\mathrm{n}\left(lab\right)+{\alpha }_7\mathrm{l}\mathrm{n}\left(lan\right)\ast \mathrm{l}\mathrm{n}\left(tlu\right)+{\alpha }_8\mathrm{l}\mathrm{n}\left(int\right)\ast \mathrm{l}\mathrm{n}\left(lab\right)\\ & +{\alpha }_9\mathrm{l}\mathrm{n}\left(int\right)\ast \mathrm{l}\mathrm{n}\left(tlu\right)+{\alpha }_{10}\mathrm{l}\mathrm{n}\left(lab\right)\ast \mathrm{l}\mathrm{n}\left(tlu\right)+{\alpha }_{11}\ln \left(liv/crp\right)\ast \mathrm{l}\mathrm{n}\left(lan\right)\\ & +{\alpha }_{12}\ln \left(liv/crp\right)\ast \mathrm{l}\mathrm{n}\left(int\right)+{\alpha }_{13}\ln \left(liv/crp\right)\ast \mathrm{l}\mathrm{n}\left(lab\right)+{\alpha }_{14}\ln \left(liv/crp\right)\\ & \ast \mathrm{l}\mathrm{n}\left(tlu\right)+v_i-v_i)\mid j\end{array}$(8)