The study was done in Karatu, Iringa rural and Njombe Town Council (TC) districts under the Agriculture Investors as Development Actors (AIDA) project. Karatu, Iringa rural and Njombe TC are designated by the Government of Tanzania as investment corridors to attract foreign agricultural investments. The procedure for selecting specific study areas is explain in detail by Ravnborg et al. (2021)². However, the selection considered agro-ecological suitability, presence of export crops, connectivity to regional markets, and presence of multiple commercial farms to provide different explanation for FAI employment seeking patterns. Proportionate stratified random sampling was used to obtain 1203 respondents of whom 397 were from Karatu, 405 from Iringa and 401 from Njombe. Calculation of the sample size is co-authored in Ravnborg et al. (2021). A structured questionnaire was used to collect information for three months. The respondents were asked if over the previous two years they had worked in FAI or non-local commercial farms. A follow up question was asked on why they had decided to seek or not to seek such employment.

Karatu district is within the Northern Investment Corridor while Iringa district and Njombe district are within the Southern Agricultural Growth Corridor of Tanzania (SAGCOT). The weather and agro-ecological characteristics of each of the districts are attractive for agricultural investments. Karatu and Njombe districts have cooler weather than Iringa district which attracted European migrants before independence. In Karatu, early European migrants invested in coffee estates on fertile volcanic soils on hills along the border with Ngorongoro conservation area. Table 1 shows the population density in Karatu district is high compare to Iringa and Njombe districts which increases pressure on fertile land. Scarcity of land for agricultural or livestock keeping activities limits employment alternatives, making agricultural wages employment an important source of livelihood to the majority of landless households. According to a survey of 399 individuals by AIDA in the area, it is a relatively less-poor population (Ravnborg et al., 2021), where 40% of the households are characterized as less-poor, 45% as poor and only 15% as very poor (Footnote 3³ and Table A1 in Appendix A elaborate how household poverty was estimated). With a population density of 106.2 person/km², poverty is more widespread in Karatu than in Iringa and Njombe.

About 40% of the households belong to the poorest households; 36% are poor and 24% are less-poor (Table 1). According to the AIDA survey, 10% of the households in the Karatu research area are landless. More than 15% have less than 1 acre of land, and 34% of the households have only between 1 and 2 acres, and 27% of the households have between 2 and 5 acres of land. Only 7% of the households have between 5 and 10 acres of land (N = 397).

The employment opportunities created in the commercial farms owned by non-locals are more numerous in Iringa and Njombe. Twelve percent of household heads take agricultural employment in these types of farms (4.3% for less than 3 months in a year, and 8% for more than 3 months in a year, respectively) (AIDA survey, N = 187 heads of household). The mean daily payment for agricultural labourers is higher for workers in commercial non-local farms than for workers in locally owned neighbouring farms, according to the AIDA survey.

Source: AIDA survey, 2019.

The mean daily pay for agricultural labourers in local farms in Karatu is 1.74 USD/day, and 2.08 USD/day at commercial, non-local farms (n = 101 local and 37 non-local, respectively).

Separate interviews with employees and non-employees adjacent to the farms revealed different working conditions and employment benefits. Some farms provide lunch, accommodation, transport services for those coming from other regions and health insurance. Others who have worked in the farm complained about delay payments of their wages while others benefitted from on job training provided. Despite low wages in each site, interviewed respondents used salary for various purposes. In Iringa, Karatu and Njombe (Figures 1-3) the respondents used wages to purchase agricultural inputs while some with contracts used salary advance to build houses and pay school fees. Non-employees benefited from increased economic activities and renting rooms or houses to workers.

From analysis point of view, use of individual motives for seeking FAI employment or not poses a challenge and unique consideration of the type of data

analytical approach. First, decision of an individual to work or not is a choice driven by several factors which are either economic or non-economic or circumstantial. Informed by previous related studies, we used statements which describe individual as either attracted or not attracted. The statements were measured as categorical and thus motives for working or not was a multilevel categorical variable because an individual might chose more than one statement. Secondly, there was lack of prior expectation as to the nature of the relation between motives and individual characteristics to verify a priori hypotheses. To address the challenges, multiple correspondence analysis (MCA) was used for the analysis of multilevel categorical responses as exploratory analysis instead of testing prior hypotheses. MCA allows simplifying complex multilevel categorical data yielding hidden patterns of data in terms of similarities which can be used to define and group together individuals with similar behaviour. Groups formed based on employment motives can further be analysed to understand conditions which are associated with household wellbeing.

MCA and K-means cluster analysis was done with the help of Statistical Package for Social Science (SPSS) Ver 24. MCA and clustering approaches are increasingly used in social sciences in detecting and identifying underlying structures in cognitive ageing and customer preferences and patterns (Bejaei, Cliff, & Singh, 2020; Costa, Santos, Cunha, Cotter, & Sousa, 2013). MCA is a non-parametric statistical tests developed for rich and complex nominal data without needing strict assumptions as compared to Chi-square analysis, Fischer’s exact test, G- statistics, and ratio test. MCA is appropriate in exploring and analysing variables in datasets which may be quantitative or qualitative, temporal or non-temporal, and/or objective or subjective. This is the case in this study where the reasons for seeking or not seeking employment in FAI farms may represent psychological, economic, socio-cultural or other forms of subjective biases, both endogenous and exogenous. MCA generated object scores which were used as inputs into k-means clustering. K-means average object scores from MCA to find centroids or means. Centroids were randomly grouped and iteratively optimized the clustering of centroids. The number of clusters was based on the one which converged to zero with a minimum number of iterations (Costa et al., 2013). Cross- tabulation was used in the preceding analysis to measure any significant association between FAI employment pattern and different segments of population and household wellbeing using Pearson Chi-square, measured at the 5% level of significance.

FAI employment opportunities promised to surrounding communities is frequently in terms of thousands or millions. These promises are used to promote FAI establishments in rural areas as pathways for rural development and contribute to reduce household poverty (Brüntrup, Absmayr, Dylla, Eckhard, & Remke, 2016; German, Cavane, Sitoe, & Braga, 2016). However, it was found that only 8% of individuals either the household head or other members of the household had worked in FAI farms. This is less than anticipated considering that FAI farm employment generating potential is expected to contribute to the majority of rural households’ welfare (Gyapong, 2020; Herrmann & Grote, 2015). FAI employment generation potential is also differ from marked statistical significance with respect to geographical locations in which 13% of individuals in Karatu more frequently take such employment than Iringa (7%) and Njombe (4%).

Karatu had more individuals who worked in FAI farms probably due to large proportion of landless households, higher population pressure and larger proportion of poorest households than in the two other districts. This coincides with findings by Ahlerup & Tengstam (2015) that people who own scanty land are frequently attracted to work in FAI farms. It implies that FAI employment generating potential is low but such employment is very important in areas with relatively many poor households, characterized with high population density and land-scarcity. However, Kareem (2018) found that high population density is not important for seeking employment in FAI farms. Therefore, FAI should not be promoted or considered as a sole provider of employment in rural areas but as an opportunity for additional income.

Table 2 shows that 92% of individuals, either the household heads or other members of the household, did not take FAI employment. The differences in the districts were statistically significant in which Njombe had 96% of individuals who not worked in FAI farms, followed by 93% in Iringa and 87% in Karatu. Individuals who had not worked in FAI were included to learn from their motives for not seeking such employment on livelihood strategies to individuals who had worked in FAI farms. Literature exists about this group of individuals which is too significantly large to ignore. Thus, motives for not seeking FAI employment were analyzed to identify patterns of seeking FAI employment and not seeking such employment.

Figure 4(a) and Figure 4(b) are two-dimension displays of individuals’ motives for seeking employment and not seeking employment respectively. The figures provide visual aid in identifying hidden structures of the data in order to

Note: ***Significant statistical association using Pearson Chi-square at 1%.

define pattern of the individuals. The length of the line or vector shows the most discriminant reasons while the short arrows mean the average or common reason for all individuals to work in FAI farms. The reasons that apply to both dimensions are at the centre of the graph. Furthermore, cross-tabulation was done between the motives for seeking and not seeking FAI employment with clusters as a second step to confirm MCA patterns.

Dimension 1 in Figure 4(a) shows lack of own land and that families losing land were the most discriminate features which defines the pattern of individuals seeking FAI farm employment and can be grouped together because they are close to each other. Along the dimension 2 only few local individuals employ others is far from other motives which suggest an outlier. But along the dimension 2 it shows there is no other employment in the area and gaining of knowledge, skills and ideas seems to be almost close to each other as one group. This indicates the pattern of individuals in this group suggests they are curious to learn through employment in FAI farms. In the middle it shows that good payment, good working conditions and provision of transport to lie close to each other which can be considered as another pattern of individuals seek FAI employment. Employment seeking pattern of these individuals was defined as preferred choice of employment. Along the dimension 2 it shows that individuals seek FAI employment because of lack of own land and losing family land as lying close to each other. This pattern was defined as individuals who seek FAI employment because of limited employment alternatives since they lack land or their family lost land or part of it land which means own farming is not an option as main source of livelihood. Close to the centre of the dimension 1 and 2 it shows that provision of health insurance provision and finding own farming too risky are common among all individuals who seek FAI employment.

Figure 4(b) present bi-plot graph which display individuals motives/reasons for not seeking FAI employment. Self-employment motive which lies along dimension 2 is very far from the rest suggesting a potential outlier in this type of analysis. Along the middle are individuals who would like to work in FAI farms but have not succeeded for various reasons is close to individuals who are not seeking FAI employment because of no or limited employment offered by existing FAI farms. The pattern of individuals in this group was defined as individuals who would like to work but not succeeded. In dimension one several motives lie close to each other which includes lack of required skills, bad payment, prefer other types of works, and having other work or employment. This pattern of individual’s reasons for not working in FAI farm was defined as Prefer self-em- ployment. Just above it are individuals who do not seek employment in FAI farms because of they don’t take paid farm works, and farms are far away. This pattern was defined as “not preferred type of work”.

Descriptions of the observations in Figure 4(a) and Figure 4(b) shows there are different causal mechanisms on individual to decide either to take or not take FAI employment which is contrary to Ahlerup & Tengstam (2015); Jann et al. (2018); Nolte, (2014) who considered only lack of land or displacement as motive for seeking FAI employment. Therefore, lack of land or limited land is a causal mechanism which in itself is not enough to make individual to decide to take or not to take employment in FAI farms. Instead this analysis found decision to work or not to work involve more than one reason or causal factors that were further analyzed to give patterns of seeking or not seeking FAI employment in the studied areas. FAI employment patterns are contextually bound to respondents and geographical areas.

Table 3 presents three types of FAI employment seeking patterns obtained from individuals motives for seeking FAI employment. Individuals who seek FAI employment due to no alternative employment, individuals who found FAI employment is a preferred type of employment and individuals who seek FAI employment for curiosity to learn new knowledge, ideas and skills in FAI farms. Overall with a marked statistical significant difference, 70% of individuals take FAI employment for curiosity to learn new knowledge, ideas and skills in FAI farms followed by 17% of individuals who work in FAI farms due to lack of alternative employment and 14% of individuals who worked in FAI farm as a preferred type of employment. The differences were statistically significant at 5% which means there is association between research location and FAI employment seeking patterns.

Furthermore, distribution of FAI employment seeking pattern due to no alternative employment vary with respect to geographical areas range from 29% in Karatu, to 6% in Njombe while in Iringa there was none. Individuals worked in FAI farms as preferred types vary in different research locations from 18% in Iringa to 14% in Karatu and 6% in Njombe. Curiosity to learn pattern for seeking FAI employment was largely among individuals from Njombe (88%), followed by 82% in Iringa and 57% in Karatu.

Therefore, distribution of FAI seeking employment is not uniform across geographical areas. Individual seek FAI employment due to no employment alternative is more frequent in areas with land availability challenges and high percentage of poverty while individuals who seek FAI employment for curiosity

Note: **significant statistical association using Pearson Chi-square at 5%. ^a4cells (44.4%) have expected count less than 5. The minimum expected count is 2.19.

to learn new is more frequent in areas with no land availability and with less poverty. Furthermore, individuals who seek FAI employment as preferred type of employment is more frequent in areas where poverty and availability of land is modest.

Table 4 shows patterns of individuals who did not seek FAI employment. Overall 100% of individual who did not seek FAI employment were self-employment followed by 0.2% who would like but not succeeded and 0.1% of individual who find FAI employment is not preferred type of work. However, the differences were not statistically significant at 5% which means there is no association between individual’s patterns for not seeking FAI employment and research locations. Therefore, self-employment is independent of a geographical area because potentially everyone prefers self-employment than being employed. Need for cash including limited livelihood options and other resources for self-employment result into seeking FAI employment. “I look for employment in FAI farms when I need money to purchase farm inputs before season begin” interview with youth in Iringa. This was also found in Karatu and Njombe where individuals seek FAI employment to meet various cash needs including school fees, hiring tractor and other social functions.

Table 5 shows distribution of gender and FAI employment seeking pattern with respect to research locations. In Karatu, 27% of men and 33% of women more frequently takes FAI employment due to lack of employment alternatives than in Njombe while Iringa there was none. However, in Njombe it was only women who worked in FAI due to lack of other employment alternatives. In Iringa 21% of women more frequently than 14% of men take FAI employment as a preferred type of employment while in Karatu it was opposite where 20% of men more frequent than 5% of women take FAI employment as a preferred type of employment. In Njombe it was a preferred type of work among men only. Men and women seeking FAI employment for curiosity to learn was observed in all research location but in varying proportions. Despite of observed differences,

Note: ^ns not statistically significant using Pearson Chi-square at 5%. ^a6 cells (66.7%) have expected count less than 5. The minimum expected count is 0.32.

association between gender and FAI employment seeking pattern was not statistically significant. Therefore FAI seeking employment pattern is independent of gender, which means FAI employment seeking pattern either due to no employment alternative or as a preferred types of employment or for curiosity to learn is not gender or location specific.

Table 6 shows distribution of individual’s age categories and FAI employment seeking patterns with respect to research locations. It was found that age categories of individuals and FAI employment seeking patterns do not have statistical significant association which means seeking FAI employment among individuals near FAI farms is independent of age. Both youth and adults seek such employment which is important for their livelihood. However, Table 6 shows a varying proportion of age categories with respect to FAI employment seeking patterns in different research locations. In Karatu it was 29% of each youth and adults frequently worked in FAI farms due to no employment alternatives while 18% of adult frequently take FAI employment as a preferred type of employment than 6% of youth. About 65% of youth frequently seek FAI employment because of curiosity to learn new knowledge, skills and ideas from FAI farms than 53% of adults. In Iringa it was similar to Karatu where 31% of adults more frequently seek FAI employment as preferred type of work than 7% of youth while 93% of youth more frequently than adults (69%) seek FAI employment for curiosity to learn new knowledge, ideas and skills in FAI farms. In Njombe it was 13% of adult more frequently seek FAI employment because of no employment alternatives

^nsNot statistically significant at 5% measured with Pearson Chi-square; ^a2 cells (33.3%) have expected count less than 5. The minimum expected count is 2.33; ^b2 cells (50%) have expected count less than 5. The minimum expected count is 2.32 computed only for a 2*2 table; ^c4 cells (66.7%) have expected count less than 5. The minimum expected count is 0.44.

^nsNot statistically significant at 5% measured with Pearson Chi-square; ^a2 cells (33.3%) have expected count less than 5. The minimum expected count is 2.88; ^b2 cells (50%) have expected count less than 5. The minimum expected count is 2.59 computed only for a 2*2 table; ^c4 cells (66.7%) have expected count less than 5. The minimum expected count is 0.44.

while youth it was none. However, 13% of youth frequently seek FAI employment as a preferred type of employment than adults. In terms of seeking FAI employment because of curiosity to learn, in Njombe it shows that both youth and adults work in FAI farm to learn new knowledge, ideas and skills.

Table 7 shows association between FAI employment pattern and household wellbeing with respect to research locations. It was found that household wellbeing as an indicator of poverty associate significantly at 5% with household wellbeing in Karatu but not in Iringa and Njombe. In Karatu it was observed that with a marked statistical significance difference in which employment pattern of individuals belonging to poorest household frequently take FAI farms employment due to lack of employment alternatives (38%) and curiosity to learn (63%) while individuals belonging to less-poor household frequently take FAI employment as a preferred type of employment.

Table 8 shows association between pattern of individual not taking FAI employment and household wellbeing with respect to research locations. It was found that individual patterns for not taking FAI employment do not associate with household wellbeing to which they belong at 5% p-value. This means if individual decide not to take FAI employment is independent of household wellbeing to which he or she belong. It implies someone from rich or poor household can decide not to take FAI employment in any area. It is not only the poor who seek or take such employment or benefit from such employment as reported in Ahlerup & Tengstam (2015).

**Statistically significant at 5% Pearson Chi-square x² = 12.34, df 2; ^nsNot statistically significant at 5% Pearson Chi-square; ^a5 cells (55.6%) have expected count less than 5. The minimum expected count is 0.41; ^b3 cells (50%) have expected count less than 5. The minimum expected count is 1.25; ^c7 cells (77.8%) have expected count less than 5. The minimum expected count is 0.19.

^nsnot statistically significant at 5% Pearson Chi-square; ^a6 cells (66.7%) have expected count less than 5. The minimum expected count is 0.26; ^bNo statistics are computed because cluster number of cases not worked is constant; ^c6 cells (66.7%) have expected count less than 5. The minimum expected count is 0.23.