Contract Farming and Gender Equity in African Landscapes (CONGENIAL)
Aug 16, 2015
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Martin Prowse, Betty Chinyamumnyamu, Ron Ngwira and Jytte Agergaard
http://www.mycongenial.com/
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Rationale
Contract farming is expanding rapidly in Africa but often
suffers from high rates of default and claims of exploitation
from smallholders
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Rationale
Contract farming is expanding rapidly in Africa but often
suffers from high rates of default and claims of exploitation
from smallholders
How can we reduce default rates?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Rationale
Farmer
Default
Sell inputs
Side-sell produce
Sell others’ produce
Firm
Default
Late supply of inputs
Purchases from spot markets
Changes in price
ACDI-VOCA (2012)
Source: Gow and Swinnen (2000)
Price (P)
Capital and reputation losses (K)
P0
P1
K1A
P-1
K-2B
Side selling
range for Farm A
Spot market
purchase range
for Firm B
P-2
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Reducing default through self-enforcing contracts
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Self-enforcing contracts
Contracts can be designed to
limit the likelihood of default
through increasing the amount
of capital and reputation
included in the scheme, thus
increasing the self-
enforcement range:
• Both parties invest in specific assets (capital)
• Name-and-shame methods (reputation)
Can including wives in contracts increase the self-enforcement range?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Rationale
Contract farming is expanding rapidly in Africa but often
suffers from high rates of default and claims of exploitation
from smallholders
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Literature from the 1970s and 1980s highlighted how contract
farming can, inter alia, contribute to:
• dependency on the firm and a form of self-exploitation in which
smallholders bear all production risk
• an intra-household distribution of labour/income that is
detrimental to women’s interests
• harmful spillover effects in local markets (e.g. food, input and
output markets)
Rationale
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Intra-household bargaining
• Wives tend to see contract farming in a favourable light if it increases
aggregate income, providing them with an incentive to co-operate in
production
• These incentives are weakened when they are not remunerated according
to labour input or when income is diverted away from household priorities
Who typically controls contract farming income?
What purposes might this income be used for?
What might the response of wives be if they are not receiving
a fair share of income?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Research Question
To what extent and how does including wives within a
contract farming scheme improve the benefits to the
firm, farms and families?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Null hypotheses
Husbands: Including wives in the contract will not be welcomed by
husbands?
Firm: Including wives in the contract has no impact on default rates?
Farms: Including wives in the contract has no impact on productivity and
the intra-household division of labour?
Families: Including wives in the contract has no impact on household
well-being?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Methodology
We combined a randomised design - where
clubs/members are randomly assigned the
intervention - with an interview schedule that
included biographical, open and closed questions
Husbands and wives interviewed separately using
the same questionnaire
Time: 18 months from October 2013
Fieldwork cash: US$12,000
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Kasungu District, Malawi We worked with AOI to evaluate the inclusion of wives
growing soya within a standard tobacco contract
Clustered randomised design
- assumed standardised effect size (Delta) @ 0.34 for
a 20% change in impact variables
- aimed for 100 clubs (50 vs. 50)
- assumed club size of 6 members
- intra-club correlation coefficient estimated at 0.22 in
Kasungu District for tobacco yields
462 households in total
227 HHs were randomly selected to receive soya
(could select multiples of 12.5kgs up to 50kgs)
235 control households
Scheme Sample size Participants % of participants Total kgs distributed
Chatoloma 39 16 41.0 225
Kasungu Central 1 75 53 70.7 1162.5
Kasungu Central 2 29 17 58.6 212.5
Mangwazu 32 28 87.5 350
Mphomwa 38 32 84.2 400
Wimbe 14 8 57.1 174
Total 227 154 66.53 2524
Huge attrition from treatment sample
Of the 227 selected for treatment, our partner, AOI
informed us that only 154 complied and only 114 planted
Why? Once bitten, twice shy - AOI distributed soya seed
to HHs in 2012/13 and deducted the cost directly from
gross tobacco proceeds
Quality of the soya seed distributed to wives was
very poor
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Implications of attrition
Reduced sample of treatment households
reduces likelihood of significant findings
Power of the RCT unlikely to reach 80%
Randomisation procedure
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Did the randomisation work? ANOVA
Sum of Squares df Mean Square F Sig.
1.11 Total members Between Groups 4.454 2 2.227 .496 .609
Within Groups 1386.389 309 4.487
Total
1390.843 311
1.2 Adult equivalent score Between Groups 1.695 2 .847 .246 .782
Within Groups 1056.972 307 3.443
Total
1058.667 309
Total land under production
2014
Between Groups .449 2 .224 .024 .976
Within Groups 2873.900 309 9.301
Total
2874.348 311
4.3 Total expenditure Between Groups 5.222E11 2 2.611E11 .461 .631
Within Groups 1.751E14 309 5.666E11
Total
1.756E14 311
3.4 Total income
Between Groups 3.745E11 2 1.873E11 .209 .812
Within Groups 2.775E14 309 8.979E11
Total 2.778E14 311
N Mean Std. Deviation Std. Error
1.11 Total members Control 173 6.70 2.108 .160
Supposed 66 6.86 2.293 .282
Planted 73 6.51 1.973 .231
Total 312 6.69 2.115 .120
1.2 Adult equivalent score Control 171 5.5498 1.81242 .13860
Supposed 66 5.6973 2.02602 .24939
Planted 73 5.4822 1.79403 .20997
Total 310 5.5653 1.85097 .10513
Total land under production 2014 Control 173 5.7881 2.87265 .21840
Supposed 66 5.8805 3.24817 .39982
Planted 73 5.7866 3.26758 .38244
Total 312 5.8073 3.04011 .17211
4.3 Total expenditure Control 173 798100.91 831163.861 63192.218
Supposed 66 693604.47 595629.634 73316.940
Planted 73 766172.15 679049.079 79476.683
Total 312 768525.38 751430.874 42541.396
3.4 Total income Control 173 760807.95 1089658.252 82845.183
Supposed 66 772151.97 800824.718 98574.709
Planted 73 682774.40 661939.763 77474.189
Total 312 744949.80 945174.348 53509.960
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Did the randomisation work?
Value df
Asymp. Sig. (2-
sided)
Pearson Chi-Square .191a 2 .909
Likelihood Ratio .190 2 .909
Linear-by-Linear
Association
.155 1 .694
N of Valid Cases 311
Value df
Asymp. Sig. (2-
sided)
Pearson Chi-Square 5.303a 6 .506
Likelihood Ratio 5.724 6 .455
Linear-by-Linear Association .702 1 .402
N of Valid Cases 311
Randomisation procedure
Logistic regression (0,1) shows asymmetric treatment
assignment in one zone – Kasungu Central 2
Implication: Control for KU2 when comparing impact
variables across T and C groups (t-tests)
Multinomial logistic regression (0, 1, 2) and Kruskal
Wallis tests showed after attrition we had an asymmetric
frequency of treatment, supposed and control
households in 4 of 6 zones
Implication: Control for spatial confounding factors when
comparing impact variables across T, S, C groups
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Did the randomisation work?
Sum of Squares df Mean Square F Sig.
5.31 Soya area in acres Between Groups 1.930 2 .965 4.464 .012
Within Groups 66.781 309 .216
Total
68.711 311
5.33 Soya production in kgs Between Groups 267442.916 2 133721.458 6.483 .002
Within Groups 3753733.219 182 20624.908
Total
4021176.135 184
5.33y Soya yield in kgs per
acre
Between Groups 553983.200 2 276991.600 7.223 .001
Within Groups 6902386.330 180 38346.591
Total
7456369.530 182
5.34 Soya % sold Between Groups 37385.977 2 18692.989 13.431 .000
Within Groups 239382.203 172 1391.757
Total
276768.180 174
5.36 Soya price per kg Between Groups 6447.450 2 3223.725 2.934 .057
Within Groups 130770.583 119 1098.912
Total
137218.033 121
N Mean Std. Deviation Std. Error
5.31 Soya area in acres Control 173 .333 .4610 .0350
Supposed 66 .371 .5156 .0635
Planted 73 .526 .4241 .0496
Total 312 .386 .4700 .0266
5.33 Soya production in kgs Control 87 178.37 143.369 15.371
Supposed 35 198.49 187.864 31.755
Planted 63 105.44 112.599 14.186
Total 185 157.34 147.832 10.869
5.33y Soya yield in kgs per acre Control 86 289.09 190.899 20.585
Supposed 35 331.06 233.636 39.492
Planted 62 189.67 178.741 22.700
Total 183 263.43 202.408 14.962
5.34 Soya % sold Control 83 61.49 38.184 4.191
Supposed 33 66.33 29.579 5.149
Planted 59 32.18 39.790 5.180
Total 175 52.52 39.883 3.015
5.36 Soya price per kg Control 66 118.26 24.984 3.075
Supposed 30 126.50 41.939 7.657
Planted 26 136.54 39.592 7.765
Total 122 124.18 33.675 3.049
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data Husbands
• AOI data on repayment / default rates
Wives
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Null hypotheses
Husbands: Including wives in the contract will not be welcomed by
husbands?
Firm: Including wives in the contract has no impact on default rates?
Farms: Including wives in the contract has no impact on productivity and
the intra-household division of labour?
Families: Including wives in the contract has no impact on household
well-being?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Wife data
Value df
Asymp. Sig. (2-
sided)
Pearson Chi-Square .205a 2 .902
Likelihood Ratio .204 2 .903
Linear-by-Linear Association .091 1 .763
N of Valid Cases 260
Value df
Asymp. Sig. (2-
sided)
Pearson Chi-Square 3.336a 2 .189
Likelihood Ratio 3.414 2 .181
Linear-by-Linear Association .232 1 .630
N of Valid Cases 260
Soya Groundnuts
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Husband farm data Soya Groundnuts
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Null hypotheses
Husbands: Including wives in the contract will not be welcomed by
husbands?
Firm: Including wives in the contract has no impact on default rates?
Farms: Including wives in the contract has no impact on productivity and
the intra-household division of labour?
Families: Including wives in the contract has no impact on household
well-being?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Wife well-being data
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Husband well-being data
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Null hypotheses
Husbands: Including wives in the contract will not be welcomed by
husbands?
Firm: Including wives in the contract has no impact on default rates?
Farms: Including wives in the contract has no impact on productivity and
the intra-household division of labour?
Families: Including wives in the contract has no impact on household
well-being?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Alliance One repayment data at club level
N Mean Std. Deviation Std. Error
Recovery Rate Control 44 79.2173 28.50410 4.29715
Supposed 17 78.5888 32.44447 7.86894
Plant 33 75.8894 32.01017 5.57226
Total 94 77.9353 30.19393 3.11427
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Null hypotheses
Husbands: Including wives in the contract will not be welcomed by
husbands?
Firm: Including wives in the contract has no impact on default rates?
Farms: Including wives in the contract has no impact on productivity and
the intra-household division of labour?
Families: Including wives in the contract has no impact on household
well-being?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Husband farm data
No significant differences for tobacco acreage, production, yield and MKW
prices
F Sig.
3.280 .041
But households that were
supposed to and did plant soya
had significantly lower $ prices for
tobacco (at 95% level)
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Husband farm data
No significant differences for maize production nor intra-household
distribution of labour
No significant differences for groundnut acreage, production, yield or price
But S and P households sold a
significantly greater % of
groundnuts (at the 90% level)
How do we interpret this?
Replacing lower soya sales?
F Sig.
2.545 .081
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Wife farm data No significant differences for tobacco acreage, production, yield and $ prices
But S and P households had
significantly lower MKW prices for
tobacco (at 90% level)
2.839 .061
F Sig.
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Wife farm data No significant differences for maize or groundnut production, nor husband
and wife labour. But significantly less child labour at 99% level…..
Sum of Squares df Mean Square F Sig.
6.14 Maize - what proportion
did children perform?
Between Groups 981.873 2 490.937 3.343 .037
Within Groups 38037.245 259 146.862
Total
39019.118 261
6.24 Tobacco - what
proportion did children
perform?
Between Groups 863.763 2 431.882 3.712 .026
Within Groups 30131.767 259 116.339
Total
30995.531 261
6.34 Soya - what proportion
did children perform?
Between Groups 652.833 2 326.416 1.306 .274
Within Groups 32480.837 130 249.853
Total
33133.669 132
6.44 Mtedza - what proportion
did children perform?
Between Groups 1255.768 2 627.884 3.283 .040
Within Groups 28302.842 148 191.235
Total
29558.609 150
N Mean Std. Deviation Std. Error
6.14 Maize - what proportion did children
perform?
Control 139 13.26 12.925 1.096
Supposed 65 9.00 11.632 1.443
Planted 58 9.91 10.534 1.383
Total 262 11.46 12.227 .755
6.24 Tobacco - what proportion did children
perform?
Control 139 10.99 11.908 1.010
Supposed 65 7.46 9.606 1.192
Planted 58 7.24 9.040 1.187
Total 262 9.29 10.898 .673
6.34 Soya - what proportion did children
perform?
Control 58 15.43 17.226 2.262
Supposed 38 10.13 15.487 2.512
Planted 37 12.78 13.634 2.241
Total 133 13.18 15.843 1.374
6.44 Mtedza - what proportion did children
perform?
Control 83 12.77 14.964 1.643
Supposed 36 5.97 11.072 1.845
Planted 32 12.66 13.499 2.386
Total 151 11.13 14.038 1.142
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Wife farm data No significant differences for maize or groundnut production, nor husband
and wife labour. But significantly less child labour at 99% level…..
Sum of Squares df Mean Square F Sig.
6.14 Maize - what proportion
did children perform?
Between Groups 981.873 2 490.937 3.343 .037
Within Groups 38037.245 259 146.862
Total
39019.118 261
6.24 Tobacco - what
proportion did children
perform?
Between Groups 863.763 2 431.882 3.712 .026
Within Groups 30131.767 259 116.339
Total
30995.531 261
6.34 Soya - what proportion
did children perform?
Between Groups 652.833 2 326.416 1.306 .274
Within Groups 32480.837 130 249.853
Total
33133.669 132
6.44 Mtedza - what proportion
did children perform?
Between Groups 1255.768 2 627.884 3.283 .040
Within Groups 28302.842 148 191.235
Total
29558.609 150
N Mean Std. Deviation Std. Error
6.14 Maize - what proportion did children
perform?
Control 139 13.26 12.925 1.096
Supposed 65 9.00 11.632 1.443
Planted 58 9.91 10.534 1.383
Total 262 11.46 12.227 .755
6.24 Tobacco - what proportion did children
perform?
Control 139 10.99 11.908 1.010
Supposed 65 7.46 9.606 1.192
Planted 58 7.24 9.040 1.187
Total 262 9.29 10.898 .673
6.34 Soya - what proportion did children
perform?
Control 58 15.43 17.226 2.262
Supposed 38 10.13 15.487 2.512
Planted 37 12.78 13.634 2.241
Total 133 13.18 15.843 1.374
6.44 Mtedza - what proportion did children
perform?
Control 83 12.77 14.964 1.643
Supposed 36 5.97 11.072 1.845
Planted 32 12.66 13.499 2.386
Total 151 11.13 14.038 1.142
But instead of running three separate ANOVAs, we need to run a
MANOVA to check to see if the combination of changes in 2+ impact
variables are a function of the intervention in question (the soya
distribution) as these variables co-vary (in other words, labour is limited)
We can do this by creating a new impact variable (dependent variable)
from the 2+ treatment variables
When doing so, we can also add control variables to account for any
confounding factors from the attrition (so we move from a MANOVA to a
MANCOVA)
Our null hypothesis is that there is no significant difference in the
changes in the amount of labour children apply to both tobacco,
groundnut and maize as a consequence of the treatment
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Fieldwork data – Wife farm data
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Descriptive Statistics
9.5 Control supposed
planted Mean Std. Deviation N
6.14 Maize - what proportion
did children perform?
Control 13.61 13.711 83
Supposed 5.69 8.956 36
Planted 11.72 11.188 32
Total 11.32 12.566 151
6.24 Tobacco - what proportion
did children perform?
Control 10.48 12.012 83
Supposed 5.14 7.220 36
Planted 7.97 9.908 32
Total 8.68 10.781 151
6.44 Mtedza - what proportion
did children perform?
Control 12.77 14.964 83
Supposed 5.97 11.072 36
Planted 12.66 13.499 32
Total 11.13 14.038 151
Effect Value F Hypothesis df Error df Sig.
The number of
observations is reduced
to 151 due to fewer
households growing
mtedza
But significance remains
at 95%
Control_supposed_planted Pillai's Trace ,086 2,154 6,000 286,000 ,048
Wilks' Lambda ,914 2,169b 6,000 284,000 ,046
Hotelling's Trace ,093 2,184 6,000 282,000 ,045
Roy's Largest Root ,083 3,959c 3,000 143,000 ,010
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Findings
Husbands and families: Including wives in the contract is welcomed by
husbands but they feel the soya intervention by AOI reduced the
household’s well-being
Firm: Including wives in the contract has no impact on default rates (but
consider the high attrition and dodgy soya seed from AOI)
Farms: Including wives in the contract led to no impact on crop
productivity but significantly reduced tobacco prices and child labour on
maize, tobacco and groundnuts
What goes around, comes around: if AOI had distributed good soya
seed in good faith their farmers and AOI would have achieved higher
prices!!
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Lessons learnt
One needs more than $12,000 to run an RCT with qualitative
components if power calculations are going to hold
Double and triple check random assignment to reduce likelihood of
spatial confounding factors
Do not trust your implementing partner to do anything right – monitor
them, evaluate their performance at every stage
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Patsogolo
Which tobacco company is going to take these results forward?
Which donor is going to fund an RCT at scale and with a trustworthy
tobacco company to investigate this further?
Which Malawian economists with expertise in RCTs would like to be part
of this impact evaluation?
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
CONGENIAL Phase II
Malawi: Double the sample. Allowing wives to choose from soya and
groundnuts
Tanzania: Double the sample. Allowing wives to choose from hybrid maize
and beans
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
In Tanzania, we are working with S&C
Ginning to evaluate the inclusion a
hybrid maize seed distribution
alongside cotton
Next steps for Tanzania Phase I….
Propensity score matching due to large
size of clubs within the clustered
randomised design
Mara Region, Tanzania
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)
Martin Prowse, Betty Chinyamumnyamu, Ron Ngwira and Jytte Agergaard
http://www.mycongenial.com/
Contract Farming and Gender Equity in
African Landscapes (CONGENIAL)