3.1. Socio- economic and demographic characteristics of respondents
Demographic, socio-economic and institutional variables are hypothesized to determine smallholder producer’s participation in contract farming in the study area. Of the total of sample households, 46% were participating in vegetable contract farming, while 54.25% were not participating in the scheme. In terms of the sex of households, 80.42% were male-headed households. From the total female-headed sample households (83), 39.76% were participating in vegetable contract farming in the study area. There is no significant difference between contract farmers and non- contract farmers in terms of sex of the household heads (Table 1). This shows that, there is no sex restriction for participation in vegetables contract farming in the study area.
Actual or perceived price is another factor that may influence the decision to join contract farming. In the 2021/22 production year, of total sample households who perceived uncertain price (193), 88.7% of the household heads participated in vegetable contract farming in order to reduce risks. A chi-square value of 243.4 test indicated that there is statistically significance difference at 1% probability level between contract participants and non-participants in terms of perception on price uncertainty. This implies that majority of smallholder farmers participate in contract farming due to fear of price uncertainty. Access to credit and other institutional services were also expected to improve smallholder farmer’s production and their welfare. The average of households’ access to credit for participants and non- participants in vegetable contract farming were 34.93% and 39.56%, respectively. The chi-square value of 8.3 tests for independence indicated that there is a significance difference between participants and non-participants at 1% probability level to the credit institutions.
The frequency of extension contacts expected to give farm households opportunities to get advisory service for their vegetable production. The average frequencies of extension contracts for the total sampled household heads were 33.96% once for every fortnight. A chi-square value of 10.4 for independence indicates that there is a statistically significant difference in the percentage of contract and non-contract farmers in terms of their frequency of extension contacts at 5% probability level. Of the total sample households about 53.3% were participated in training related to vegetable production. A chi-test result indicated that there is a statistically significant difference at 1% probability level between contract participants and non- participants in terms of their training attendance.
Access to market information enables smallholder farmers to search for and associate information available for different market channels to manage the cost benefit analysis and related factors in vegetable contract farming in the study area. Of the total sampled household heads, 84.2% had access to market information and of these 75.2%% participated in vegetable contract farming. A chi-squared test shows there is statistically significance difference and 5% probability level in the percentage of participant and non-participants vegetable contract farming in terms of access to market information.
Table 1
Descriptive statistics of dummy variables
Variables
|
Response
|
Total
|
Non- participant
|
Participant
|
X2
Value
|
(N = 424)
|
(N = 230)
|
(N = 194
|
Freq.
|
%
|
Freq.
|
%
|
Freq.
|
%
|
Sex of Household Head
|
Male Headed (1)
|
341
|
80.42
|
180
|
52.79
|
161
|
47.21
|
1.495
|
Female Headed (0)
|
83
|
19.58
|
50
|
60.24
|
33
|
39.76
|
Education level of Household Head
|
Illiterate (1)
|
47
|
11.08
|
24
|
51.06
|
23
|
48.94
|
1.172
|
Read and Write (2)
|
130
|
30.66
|
74
|
56.92
|
56
|
43.08
|
Primary school (1–8)
|
75
|
17.69
|
38
|
50.67
|
37
|
49.33
|
Secondary school (9–12)
|
109
|
25.71
|
61
|
55.96
|
48
|
44.04
|
Higher education
|
63
|
14.86
|
33
|
52.38
|
30
|
47.62
|
Perception about price uncertainty
|
Yes (1)
|
193
|
45.52
|
25
|
10.87
|
168
|
88.7
|
243.36***
|
No (0)
|
231
|
54.48
|
205
|
89.13
|
26
|
11.3
|
Access to Credit
|
Yes (1)
|
142
|
33.49
|
91
|
39.56
|
51
|
34.93
|
8.328***
|
No (0)
|
282
|
66.51
|
139
|
60.44
|
143
|
51.44
|
Frequency Extension Contacts
|
Weekly (1)
|
115
|
27.12
|
61
|
26.5
|
54
|
23.5
|
10.41**
|
Fortnight(2)
|
144
|
33.96
|
74
|
32.2
|
70
|
30.4
|
Monthly (3)
|
81
|
19.10
|
37
|
16.1
|
44
|
19.1
|
Others (4)
|
4
|
0.94
|
3
|
1.3
|
1
|
1.7
|
Never (5)
|
80
|
18.87
|
55
|
23.9
|
25
|
10.9
|
Training on vegetable farming
|
Yes (1)
|
226
|
53.3
|
59
|
25.7
|
167
|
72.6
|
154.39***
|
No (0)
|
198
|
46.7
|
171
|
74.3
|
27
|
27.4
|
Membership in saving group ‘Equib’
|
Yes (1)
|
162
|
38.21
|
146
|
63.5
|
78
|
33.9
|
0.61
|
No (0)
|
262
|
61.79
|
84
|
36.5
|
116
|
62.1
|
Market information
|
Yes (1)
|
357
|
84.2
|
184
|
80
|
173
|
75.2
|
6.659**
|
No (0)
|
67
|
15.8
|
46
|
20
|
21
|
24.8
|
Disease and pest infestation
|
Yes (1)
|
411
|
96.93
|
223
|
97
|
188
|
81.7
|
0.0009
|
No (0)
|
13
|
3.07
|
7
|
3
|
6
|
18.3
|
Note: *** and ** represent level of significance at 1% and 5%respectively |
Source: Survey Data (2022) |
In the 2021/22 production year, the average experiences of household heads in vegetable farming were 10. 8 years. The mean of smallholder’s farm experience for participants and non-participants in vegetable farming were 11.35 and 10.36 years, respectively. The t-test result show that, there is a statistically significance mean difference between the two groups in terms of their farm experience at 10% significance level.
The total average value of asset owned by sampled household heads from off and non- farm income sources in the study area were 55,128.95 ETB for the whole sample; this was estimated at 52,258.6 and 57,550 ETB for participants and non- participants respectively. The t-test for equality of means for household heads income from off-farm and non-farm income sources among two groups are statistically significant at 5% probability level.
The area average of total farm land allocated to vegetable crops in 2021/22 of the entire sampled households was 1.24 hectares with 1.25ha and 1.23ha for participants and non-participants. The t-test result shows that, there is a statistically significance mean differences between the two groups in terms of the proportion of total farm land to vegetable farming at 1% probability level.
Table 2
Descriptive statistics of continues variables
Variables
|
Total
|
Non-participant
|
Participant
|
t-test
|
(N = 424)
|
(N = 230)
|
(194)
|
Mean
|
Std. Dev
|
Mean
|
Std. Dev
|
Mean
|
Std. Dev
|
Age of Household Head
|
41.099
|
6.514
|
40.639
|
5.938
|
41.644
|
7.114
|
-1.586
|
Household family size
|
4.349
|
1.697
|
4.422
|
1.834
|
4.263
|
1.519
|
0.96
|
Dependency ratios
|
1.023
|
0.711
|
1.045
|
0.678
|
0.997
|
0.749
|
0.688
|
Experience of Household Head
|
10.8
|
5.84
|
10.36
|
5.75
|
11.35
|
5.9
|
-1.74*
|
Off- farm and non – farm income
|
55129
|
26936.6
|
57550
|
28896.7
|
52258.6
|
24170.6
|
2.02**
|
Proportion of vegetable to total farm land
|
1.24
|
0.027
|
1.23
|
0.04
|
1.25
|
0.036
|
-2.78***
|
Distance from Market center
|
6.507
|
1.982
|
6.435
|
2.07
|
6.59
|
1.875
|
-0.818
|
Note: ***, ** and * represent level of significance at1%, 5%ansd 10% respectively |
Source: Survey data (2022) |
3.2. Propensity score matching results
The probit regression model was used to estimate the propensity score for participant and non- participant households in contract farming. The pre-intervention variables were taken as explanatory variables and assumed to affect the participation in vegetables contract farming. Before proceeding to the impact estimation, the variance inflation factor (VIF) was applied to test for the presence of strong multicollinearity problem among explanatory variables. There was no serious problem of multicollinearity and hence no explanatory variable was dropped from the estimated model. Similarly, Breusch- pagan test for heteroscedensity was used to check the existence of heteroscedasticity of the variance and there was no heteroscedasticity problem in the model.
The estimated probit regression model (Table 3) appears to perform well for the intended matching exercise. The pseudo- R2 value is 0.0962, low R2 value shows that participants households do not have much distinct characteristics overall and implies there is a good match between contract participants and non- participants. The interest of the matching producer is to get households from vegetable producer in non-participants contract farming with similar probability of participants in a contract farming given explanatory variables. Probit model was used to calculate propensity scores by running psmatch2 command.
Table 3
Probit regression estimates of vegetable CF participation on HH income (Psmatch2)
Contract farming
|
Coef.
|
Std. Err.
|
Z
|
Dy/dx
|
Age of household head
|
0.0097
|
0.016
|
0.60
|
0.0038
|
Sex of household head
|
-0.343
|
0.223
|
-1.54
|
-0.136
|
Education level of Household Head
|
-0.045
|
0.048
|
-0.94
|
-0.018
|
Dependency ratios
|
-0.116
|
0.103
|
-1.13
|
-0.046
|
Experience of household head
|
0.006
|
0.016
|
0.34
|
0.0022
|
Off- farm and non – farm income
|
-0.000
|
3.12e
|
0.6
|
-4.53e-06
|
Proportion of vegetable to total farm land
|
0.279***
|
0.08
|
3.48
|
0.1104
|
Access to Credit
|
-0.585***
|
0.152
|
-3.84
|
-0.2246
|
Frequency extension contacts
|
-0.177***
|
0.055
|
-3.22
|
-0.0702
|
Member of saving group ‘Equib’
|
0.039
|
0.142
|
0.27
|
0.0153
|
Market information
|
0.664**
|
0.226
|
2.94
|
0.245
|
Distance of the market center
|
0.061*
|
0.034
|
1.80
|
0.024
|
Disease and pest
|
-0.003
|
0.4
|
-0.01
|
-0.001
|
_cons
|
-0.424
|
0.845
|
-0.50
|
|
Log likelihood = -264.24758
|
Pseudo R2 = 0.0962
|
|
|
|
LR chi2 (13) = 56.23***
|
Number of obs. = 424
|
|
|
|
Note ***, **, * represents the level of significant at 1%, 5% and 10%, respectively
Source: Survey Data (2022)
If the numbers of explanatory variables affecting the participation decision are limited, it created a good opportunity for matching and it makes the matching producer less difficult since matching algorism is implemented to estimate significant differences of explanatory variables between participants and non-participant groups. The maximum likelihood estimates of the probit regression model result shows that, five variables out of thirteen variables were significant and affect the participation of smallholder farmers in vegetables contract farming.
Before proceeded to the next steps, it is better to see the common support region that was imposed (the second steps of PSM). The common support helps to check to identify the region of common support between the treatment (participants) and comparison groups (non-participants). In the literature review, different ways were suggested to analysis the common support region. The most used one is visual analysis of the density distribution of the propensity scores for both groups.
To choose the best matching algorithm, the most commonly used methods are nearest neighbor, kernel and radius caliper matching methods. To select the best matching algorithms, different criteria are recommended and used by different scholars. As Dehejia and Wahba (2002) states, the equal means test referred as balancing test, pseudo- R2, Mean Standard Bias and matched sample size are recommended as best criteria to selected best common support region in propensity matching scores. Table 5 shows all matching algorithms were undertaken and offer the same results. The caliper radius matching with radius (0.1) has relatively low pseudo with best balancing test (all explanatory variables insignificant) and large sample size as compared to the other alternatives in both outcome variables.
Table 5
Performance of the different matching algorithms on Household income
Matching Estimator
|
Matching performance criteria
|
Balancing
Test*
|
Pseudo R2
|
Mean Standard Bias
|
Matched sample size
|
Nearest Neighbor
|
Neighbor (1)
|
13
|
0.031
|
8.4
|
416
|
Neighbor(2)
|
13
|
0.018
|
6.6
|
416
|
Neighbor(3)
|
13
|
0.011
|
6.1
|
416
|
Neighbor(4)
|
13
|
0.012
|
5.1
|
416
|
Neighbor(5)
|
13
|
0.008
|
4.7
|
416
|
Kernel
|
Bwidth (0.1)
|
13
|
0.031
|
8.4
|
416
|
Bwidth (0.25)
|
13
|
0.031
|
8.4
|
416
|
Bwidth (0.5)
|
13
|
0.031
|
8.4
|
416
|
Caliper or Radius
|
Radius caliper (0.1)
|
13
|
0.005
|
4.1
|
416
|
Radius caliper (0.25)
|
13
|
0.022
|
6.5
|
416
|
Radius caliper (0.5)
|
13
|
0.079
|
13.1
|
416
|
* Number of independent variables with no statistically significant mean difference between the matched groups of households |
Source: Survey Data (2022) |
The third stage is to conduct the balancing test to know whether there is statistically significant difference in the mean value of the two groups of the sampled households. It is better to compare the influence of the background characteristics of both treated and comparison groups before matching for variable selection. Table 6 shows that, the t-test of covariate balance test resulted in statistically insignificant difference between treated and comparison groups in selected variables.
Table 6
Covariate balance test for the impact of participation CF on HH income.
Variable
|
Mean % reduction
|
t-test
|
V(T)/V(C)
|
Treated
|
Control
|
%bias
|
t
|
p > t
|
Age of household head
|
41.64
|
40.98
|
10.1
|
1.00
|
0.320
|
1.43*
|
Sex of household head
|
0.83
|
0.83
|
0.2
|
0.02
|
0.984
|
.
|
Education level of Household head
|
3.36
|
3.39
|
-2.7
|
-0.27
|
0.787
|
1.08
|
Dependency ratio
|
0.997
|
0.98
|
3.0
|
0.30
|
0.762
|
1.41*
|
Experience of household head
|
11.35
|
10.94
|
7.0
|
0.67
|
0.502
|
0.94
|
Off- farm and non – farm income
|
52259
|
53099
|
-3.2
|
-0.37
|
0.715
|
1.32
|
Proportion of vegetable to total land
|
2.63
|
2.61
|
2.2
|
0.23
|
0.818
|
0.89
|
Access to credit
|
0.26
|
0.28
|
-4.5
|
-0.47
|
0.642
|
.
|
Frequency extension contacts
|
2.35
|
2.3055
|
2.9
|
0.30
|
0.764
|
0.85
|
Member of saving group ‘Equib’
|
0.40
|
0.41607
|
-2.9
|
-0.28
|
0.780
|
.
|
Market information
|
0.89
|
0.8715
|
5.6
|
0.62
|
0.538
|
.
|
Distance of the market center
|
6.59
|
6.6654
|
-3.7
|
-0.36
|
0.722
|
0.77
|
Disease and pest
|
0.97
|
0.97742
|
-4.8
|
-0.51
|
0.611
|
.
|
Source: Own computation result based on the survey data (2022)
Once the difference between the outcomes of participants in contract farming and non- participants was computed, the next stage is to provide evidences of the impact of participation in vegetable contract farming on the household’s incomes (Table 7). After controlling the pre-participation differences, we found out that participation in contract farming has decreased average income of the participant by 2,941.95 ETB in 2021/22 production year. This result shows that, participating in contract farming decreases the household’s gross annual income by 3.8%. This implies that, contract farming is not a panacea that benefits smallholder farmers in different contexts. Information from key informant interview and focus group discussion confirms that vegetable producers were willing to enter into contract farming in order to access production inputs, advice on market specification and production management and to reduce risks and uncertainty. This finding is consistent with the finding of Olounlade et al. (2020) Who found out that, smallholder participation in contract farming lower the income of farmers in rural Benin. Similarly, Abdulai and Al-hassan (2016) Reported that participation in soyabean contract farming in Ghana decreased the average annual income of farmers and Seerp, (2018) reported similar finding in his study comparing participation in contract farming and out- grower scheme in Wageningen, the Netherland. On the other hand, Gemechu et al (2017) and Addisu et al (2020) found positive income effect of participation in contract farming in Ethiopian context. In general, the finding is inconclusive and mixed. Contract farming is still important institutional arrangement for linking smallholder high value crop producers to both output and input market. However, whether smallholder commodity producer’s income would change as a result of participation in contract farming depends on context and the nature of contract farming and the way it is implemented.
Table 7
ATT effect of vegetable contract farming on the household‘s income.
Variable Sample
|
Treated
|
Controls
|
Difference
|
S.E.
|
T-stat
|
Gross Annual income unmatched
|
75,755.1
|
87,942.5
|
-12,187.4
|
3,710.2
|
-3.28***
|
ATT
|
76,525.9
|
79,467.8
|
-2, 941.95
|
4,626.95
|
-0.64
|
Source: Own computation result based on the survey data (2022)
Finally, sensitivity test was computed to check the robustness of the estimation covariates to show whether the hidden bias affects the estimated ATT or not. Therefore, a sensitivity test was used to investigate whether the causal effect estimated from the PSM is susceptible to the influence of the unobservable covariates. Table 8 shows the sensitivity analysis of hidden bias for the impact of vegetable contact farming on household’s income. To check for unobservable biases Rosenbaum bounding approaches were used. As reported in Table 8, the inference for the effect of vegetable contract farming is not changing, though the participant and non-participant households have been allowed to differ their odds of being treated up to gamma = 5 in terms of unobservable covariates. This shows all outcome variables estimated at various levels of critical values of gamma and p- values are significant.
Table 8
Sensitivity analysis of hidden bias for the impact of CF on the household‘s income
Gamma
|
sig+
|
sig-
|
1
|
.000861
|
.000861
|
2
|
3.1e-14
|
.804253
|
3
|
0
|
.999334
|
4
|
0
|
1
|
5
|
0
|
1
|
*gamma - log odds of differential assignment due to unobserved factors
sig+ - upper bound significance level
sig- - lower bound significance level
|