This chapter presents empirical results obtained from examination of the various research questions posed in Chapter 2.
7.1.1 Estimation Procedure
Profitability is defined as net revenue (total revenue total cost) per unit of output, or rupees per egg in the case of layers and rupees per bird in the case of broilers. This profitability is calculated for all 320 sample units using the relevant information on revenue, cost, and output in terms of quantity.
Next, profitability across small farms and large farms is averaged separately, and the difference between the two are statistically tested for confirming the hypotheses presented in Chapter 2.
7.1.2 Results
Table 7.1 and Figure 7.1 display the distribution of layer units by profitability. A careful inspection of these data shows three interesting points. First, layer units, on the whole, do not appear to be performing well. For instance, 11 out of 161 sample units are found to be net losers. Second, most of the units with positive profitability are found to cluster in the range of 10 paise to 40 paise per egg. Third, there is an almost equal percentage distribution of both small and large layer units across all the ranges of profitability, making it difficult to say whether small or large units earn more profit per unit of output. Fourth, the average profit of first quantile (with mean output of 6,676 eggs) is highest, as compared together quantiles with higher mean output.
Table 7.2 reports these results for the broiler units. Looking at the data arrayed in Table 7.2, one notices that broiler units, on the whole, tend to be profitable, unlike layer units. Only three out of 158 units tend to be operating at a loss or with zero profit; the remaining 155 units are shown to be making a profit (i.e., have positive values). It is again difficult to determine from these data whether small or large units are making more profit per unit of output, since the distribution of the units seems to be spread evenly across all the ranges.
The estimated mean values of profitability for the small and large farms are reported in Figure 7.2, Figure 7.3, and Table 7.3, along with F and t test results.
The tests of equality of means in the case of small versus large layers show that the null hypothesis () is rejected, and the alternate hypothesis () is accepted. It is difficult, however, to establish from this with great confidence that small layers have higher average profitability than large layers, because the null hypothesis () against () is not rejected at five percent level of significance. The test of equality of means of large versus small broilers is not statistically rejected. Hence it is difficult to establish the hypothesis that smallscale producers have greater profit (average) per unit of output than do largescale producers.
7.2.1 Empirical Procedures
Do small producers expend more effort on and make more of an investment in pollution abatement than do largescale producers? Conventional wisdom suggests that largescale producers may be worse offenders than small producers. For instance, if many smallscale producers are concentrated in the same place and do not use abatement procedures, then even smallscale producers may be the worst polluters. And smallholders may pollute the water table more due to dumping of poultry manure. Thus conventional wisdom may require empirical testing for confirmation.
The procedure adopted to conduct such testing is to regress environmental cost per unit of output (envteste) with a select list of the explanatory variables. The former is defined as the sum of the costs of controlling flies + dead bird disposal + cost of pollution payment + manure disposal cost of value (including imputed) and of manure used (consumed). The explanatory variables selected here include family labour (family), number of houses within a 500meter radius (proxy for concentration of units in small area) House 500, total number of years of experience  Exptotal, information source (Infortvn), Independent/Contract farmers Dummy (Idcontract), Decgender= Gender of Decisionmaker, State dummy, education, and age of the decisionmakers (Educdm and Decage). The regression is run with an intercept dummy (i.e., Scale) scale, which takes a value of 1 for small, and 0 for large. If the intercept dummy has a positive value, then it confirms that small producers invest a larger amount in pollution abatement than do large producers (after removing the impact of independent variables on environmental cost). Alternatively, if the intercept takes negative values, then it is the reverse.
7.2.2 Empirical Results
The estimated regression results of pollution abatement costs for layers and broilers are reported in Table 7.4. An inspection of the regression results reported in Table 7.4 shows that in both the layer and broiler equations the scale coefficient is positive. It is not statistically significant in the case of layers, but is very significant in the case of broilers. Thus it confirms, at least in the case of broilers, that smallscale producers tend to spend a larger amount per unit of output in terms of pollution abatement than the largescale producers tend to do.
7.3.1 Empirical Procedure
The investigations described thus far have attempted to determine the profitability of small versus large producers as well as the significance of different variables/factors in explaining pollution abatement costs. The detailed examinations conducted did show some difference in the average profitability of small producers visàvis large producers, but the difference was not large enough to pass the test of statistical significance. Furthermore, pollution abatement costs did affect profitability, especially that of broilers, and affected the profitability of small producers more than large producers. The question that now needs to be researched is whether small or large producers are more efficient. The methodology employed to answer this question involves estimating frontier profit function and explaining the differential performance of sample farms (away from the frontier line) in terms of differences in transaction costs and pollution abatement costs. As a first step towards estimating frontier profit function, the dependent variable, profitability (PRi = TRi TVCi) where Pri = Profit in Rs. of ith unit, TRi = Total Revenue of ith unit, TVCi = Total Variable Cost of ith unit, is regressed with frontier variables. The explanatory variables of the frontier are:
PFDOC 
= price of day old chicks 
PLB 
= wage rate (of male workers) 
PF 
= price of feeds 
PSH 
= price of output (eggs or broilers) 
FLAB 
= family labour per unit of output 
VK 
= value of capital stock per unit of output 
PHLB 
= labour housing, slope dummy of PLB  A large number of workers have been provided with houses. Ideally, the rental value of the houses should be included in the wage rate variable, PLB. We do not have the rental value of these houses. In case this rental value is not included in PLB, the observation of PLB (provided with houses) is downward biased. To capture these downward biased (errors in observations), we included a slope dummy for PLB. The slope Dummy has value = 1 if labourer is provided accommodation, or = 0, otherwise. 
FCR 
= Feed Conversion Ratio (Amount of feed used per unit of output, i.e., Feed/Egg or Feed/Bird (for broilers) it is a variable to capture technological change. 
SCALE DUMMY 
= 1 for small units = 0 for large units 
The OLS regression equation may be written in log form as
PRi = a_{o} + a_{1}LNPFDOC + a_{2}LNPLB + a_{3}LNPF + a_{4}LNPSH + a_{5}LNFLB/Q + a_{6}LNVK/Q + a_{7}LNPHLB + a_{8}LN(PLB*FLB) + a_{9}(SCALE DUMMY*PLB) + a_{10} LNFCR + a_{11}.SCALE DUMMY
Where Q represents total output of unit.
We expect profitability to be inversely related to price of inputs LNPFDOC (price of day old chicks), LNPLB (wage rate), LNPF (price of feeds), and FCR (Feed Conversion Ratio); and positively related to output price and factor inputs LNPSH (price of eggs/broiler), LNFLAB (family labour per unit of output), LNVK (value of capital stock per unit of output).
The next step is to run a second regression, where the residual terms obtained from the application of the first regression equation, also referred to as technical inefficiency (technical inefficiency ranges from 0, i.e., the least efficient to 1, i.e., the most efficient), is the dependent variable, and the explanatory variables are transaction costs, and pollution abatements costs. That is, technical efficiency/inefficiency across farms are sought to be explained in terms of differences in transaction costs, and pollution abatement costs. The proxy variables selected for transaction costs are: age of the decisionmaker (in years)  DECAGE; education of the decisionmaker (dummy variable =1 for secondary and above) EDUCDM; information source (dummy variable =1 for radio, TV, Newspaper) INFORTVN; distance to output market OUTMARKDIS; gender of decision maker (dummy variable, male = 1) decgender; access to credit (dummy = 1 if credit obtained) CREDITDUM; HHH Primary source of income (Dummy=0 if poultry is primary source) HHOO; whether member of community organization or not COMORGYN; years of experience in poultry EXPTOTAL; No. of training programmes attended by the poultry owner POUPROGNUM; and Region/state characteristics (Dummy = 1 for Haryana and 0 for Andhra Pradesh) REGDUM. Pollution abatement costs (ENVTOSTQ) are taken as costs of controlling flies + dead bird disposal cost + cost of pollution payment + manure disposal cost + value (included imputed) of manure used/consumed.
From the estimated d coefficients of the second regression (using technical inefficiency as the dependent variable), one can make inferences about the direction and magnitude of the contribution of each determinant to the relative inefficiency of the farm in question. A significant positive coefficient means a positive contribution to increased inefficiency and vice versa.
7.3.2 Results
The estimated regression results of frontier function for layer^{[63]} and broiler are separately displayed in Tables 7.5. and 7.6, respectively.
The estimated results reported in Table 7.5 show profitability to be negatively related to the price of chicks and the price of feeds, and positively related to the price of eggs. The coefficients of LNFLAB (family labour per unit of output), LNVK (value of capital stock), and LNPHLBC (slope of labour housing) are found to be not statistically significant.
Similarly, the estimated results reported in Table 7.6 for broilers show profitability to be negatively correlated to the price of chicks, the price of labour (i.e., wage rate), and the price of feeds; and positively related to the price of broilers. All of these coefficients are found to be statistically significant. The coefficient of LNFLAB (family labour per unit of output) is found to be not statistically significant. A point worth noting is that the technological variable the Feed Conversion Ratio (FCR) is found to be statistically significant and with the expected sign, i.e., the higher the FCR, the lower the profitability and vice versa. This contrasts with the results received for layers, where FCR was not found to be significant. Again, in the case of broilers, the small units are relatively more profitable than the large ones as the scale dummy coefficient is positive and significant. The second step regression results (inefficiency effects due to transaction costs, and pollution abatement costs) for layers and broilers are displayed in Tables 7.7 and 7.8, respectively.
The estimated d coefficient (inefficiency effects) displayed in Table 7.7 show variables, namely DECAGE (age of decision maker), REGDUM (regional character), and ENVTOSTQ (pollution abatement costs), as being statistically significant in explaining inefficiency effects across farms ( is 0.994179, i.e., close to 1, implying that the inefficiency effects are important in explaining profits across farms). Thus one can see from a careful inspection of this table that inefficiency effects are inversely related to the age of the decisionmaker (i.e., the higher the age of the decisionmaker, the lower the inefficiency effects and vice versa) and regional/state character (the more the state imposed distortion, the higher the inefficiency effects and vice versa). These data can be interpreted as saying that transaction costs (proxies by variables such as DECAGE) play a significant role in explaining the technical inefficiency of layer farms. In addition to transaction costs, another significant explanatory variable that explains farm inefficiency is pollution abatement costs. The coefficient of ENVTOSTQ is positive and statistically significant in this result, which is based on the pooled observation of small and large units.
The estimated d coefficient (inefficiency effects) displayed in Table 7.8 show that regional/state character (the more the state imposed distortion, the higher the inefficiency effects and vice versa) had a significant impact. Unlike in the case of layers, however, pollution abatement cost is not a significant variable in explaining the inefficiency effects of broilers. The coefficient of ENVTOSTQ is statistically insignificant when the data of all sample units are pooled.
How do transaction costs and pollution abatement costs help to explain the comparative inefficiency of small versus large farms? Alternatively, do the transaction costs and pollution costs play a significantly greater role in explaining the technical inefficiency of the small farm than it does for the large farm? To test this the second step regression was reestimated separately for small and large farms. The estimated results for layers and broilers are displayed in Tables 7.9 and 7.10, respectively.
First, take a look at the results displayed in Table 7.9 for layers. A careful inspection of these results show that both transaction costs and pollution abatement costs significantly influence the efficiency of the large farm visàvis the small farm. The coefficients of DECAGE, REGDUM, OUTMARKDIS (all proxies for transaction costs) and ENVTOSTQ (pollution abatement costs) are all statistically significant. Next, look at the results reported in Table 7.10 for broilers. A glance at these results shows that the efficiency of smallscale broiler producers is influenced by both transaction costs and stateimposedpolicy distortions: the coefficient REGDOM and EDUCDM are significant.
For largescale broiler producers, the most significant variables affecting efficiency are ENVTOSTQ and INFORTVN.
7.4.1 Empirical Procedure
A notable feature of the structural change taking place in the Indian poultry industry is the rapid increase in contract farming. Contract farming first made its appearance in layers more than two decades ago, primarily to meet the demand for manufacturing of egg powder. The export demand for egg powder was high and, due to rising demand, the number of egg powder plants also increased. At the same time, contract farming in layers also started to flourish. As the E.U. and other western countries started instituting new residue limits for pesticides, however, and insisted for submission of execution of RMP, exports of egg powder started to drastically decline. That led to the closure of many egg powder plants and, subsequently, contract farming in layers also came to a near halt.
During the last few years contract farming has started to grow once again, but this time in broilers. It is growing rapidly in the southern and northern regions of India, where several poultry houses had been lying idle. Owners of such idle poultry farms had no funds for farm operation and found it better to sign up as broiler growers for a fixed fee. The integrator that signed up these poultry growers supplied DOCs and, in most cases, feed, vaccines, and other inputs. The birds are bought back by the integrator for his processing plant or retail trade. For the integrator, the sizeable volumes provide the advantage of lower production costs and, for the farmers, a revenue is guaranteed from assets that had been idle. The contract between the integrator and the broiler growers remains very loose in the sense that the agreement does not legally bind the parties, who operate mainly on mutual trust and confidence.
The question that now needs to be answered pertains to the comparative performance, in terms of financial profit, of independent and contract farms. To determine this, it is necessary to first sum up and draw average profitability separately for the sample population of contract farms and independent farms. The sample contains 23 contract broiler farms (17 smallsize farms and six largesize farms) and 136 independent broiler farms (93 smallsize farms and 43 largesize farms). Hence we calculate profitability of contract farms by averaging across profitability of individual units, and reporting the same excise for independent farms. Next, we test the significance of the difference in the profitability between independent and contract farms through statistical methods to determine whether the differences are really significant or not.
7.4.2 Empirical Results
Table 7.11 displays the calculations of the average profitability of contract farms and independent farms, both large and small. The results of the calculations are also exhibited in Figure 7.4.
A glance at these results shows that, in terms of financial profitability, independent farms on the whole perform better than contract farms. Average profitability in the case of the total number of independent broiler farms works out to Rs. 12.43 per bird compared with Rs. 1.62 per bird for contract farms. Not only are the differences substantial, they are statistically significant as well. When the comparison is drawn between small independent farms and small contract farms, or between large independent and large contract farms, the differences remain substantial and statistically significant enough to prove the prevailing contention that independent farms are more profitable, on the whole, than contract farms.
When comparisons are drawn, however, between small and large farms within the same category (i.e., small contract farm vs. large contract farm, or small independent farm vs. large independent farm), then the differences are not significant enough to state categorically that small farms are more profitable than large farms.
Table 7.1: Distribution of Layer Units by Profitability (Rs./egg) 

Profitability (Rs./egg) 
(<10,000 birds) 
(>10,000 birds) 
Total Layer 
1.00 0.00 
4 
7 
11 
0.00 0.10 
11 
18 
29 
0.10 0.20 
12 
24 
36 
0.20 0.30 
8 
32 
40 
0.30 0.40 
23 
15 
38 
0.40 0.48 
5 
2 
7 
Total 
63 
98 
161 
Note: Average price of output is Rs. 1.19 per egg.
Source of Data: Indian Poultry Survey, 2002
Table 7.2: Distribution of Broiler Units* by Profitability (Rs./bird) 

Profitability 
(<10,000 birds) 
(>10,000 birds) 
Total Units (No.) 
0.50 0.00 
2 
1 
3 
0.00 1.00 
11 
2 
13 
1.00 2.00 
8 
5 
13 
2.00 3.00 
5 
7 
12 
3.00 5.00 
10 
7 
17 
5.00 7.00 
10 
4 
14 
7.00 11.00 
19 
6 
25 
11.00 15.00 
18 
3 
21 
15.00 20.00 
10 
7 
17 
20.00 30.00 
7 
4 
11 
30.00 53.00 
9 
3 
12 
Total Nos. 
109 
49 
158 
Note: Average price of Broiler is Rs. 29.45 per Kg.
* Excluding one observation which had extremely high negative value (probably due to measurement error in collection of data of a variable)
Source of Data: Indian Poultry Survey, 2002
Table 7.3: Average Profitability of Small and Large Farms: Statistical Test for Equal Means of Two Samples (for layers: Rs./egg, and for broiler: Rs./bird) 

Category 
Sample 
Mean value () 
Equality of averages: Large vs. Small 
tstatistics 
Degree of Freedom (df) 
sig (2 tailed test) 

Layer: 
Small (LS) 
63 
0.226 


Large (LL) 
98 
0.168 
0.536 
2.011 
159 
0.046 

All 
161 
0.200 


Broiler: 
Small (BS) 
109 
11.36 


Large (BL) 
49 
9.98 
0.624 
0.77 
156 
0.44 

All 

158 
10.93 

Notes: i) Small = Less than 10,000 birds.
ii) Large = More than 10,000 birds
iii) Levene s test for equality of variance shows equal variances for small and large sample.
iv) Results based on computer package: SPSS
Source of Data: Indian Poultry Survey, 2002.
Table 7.4: Regression Results of Pollution Abatement Cost 

Independent variables 
Coefficients 

Layers: Dependent variable: envteste 

Scale 
n.s. 

Family lab 
n.s. 

Houses 500 
n.s. 

Exptotal 
n.s. 

Infortvn 
n.s. 

State dummy 
0.01451 

Educdm 
n.s. 

Decage 
n.s. 

Decgender^{1} 
n.s. 

Constant 
0.02011 

R^{2} = 0.404 N = 161 

Broilers: Dependent Variable: envteste 

Independent variables 
Coefficients 

Idcontract 
n.s. 

Scale 
1.41392 

Family lab 
n.s. 

Houses 500 
0.00511 

Exptotal 
n.s. 

Infortvn 
n.s. 

state dummy 
1.51862 

Educdm 
n.s. 

Decage 
n.s. 

Decgender 
n.s. 

Constant 
3.12040 


R^{2} = 0.306 
N = 154 
Regression results are based on computer package STRATA. 
Source of Data: Indian Poultry Survey, 2002
Table 7.5: Profitability of Indian Layers: Regression Results of Frontier Function 

Dependent Variable: Profit/Egg 

Independent Variable 
Description 
Estimated Coefficients 
LNPFDOC 
 Price of chicks 
0.47 
LNPLB 
 Wage Rate 
n.s. 
LNPF 
 Price of feeds 
1.46 
LNPSH 
 Price of Egg 
0.93 
LNFLAB 
 Family labour/Q 
n.s. 
LNVK 
 Value of capital stock/Q 
n.s. 
LNPHLB 
 Slope of labour housing 
n.s. 
LN (PLB*FLAB) 
 Wage Rate x Family Labour 
n.s. 
LN (SCALE DUMMY*PLB) 
 Wage Rate * SCALE DUMMY 
0.60 
FCR 
 Feed Conversion Ratio 
n.s. 
SCALE DUMMY 
 Small Units 
4.34 
Constt. 

10.64 
Notes: No. of observations (Pooled data of small and large units): 158,
Mean Efficiency: 0.775
n.s.= not significant
Regression results are based on computer package FRONTIER IV
Source of Data: Indian Poultry Survey, 2002
Table 7. 6: Profitability of Indian Broilers: Regression Results of Frontier Function* 

Where the Dependent Variable: Profit/Bird 

Independent Variable 
Description 
Estimated Coefficients 
LNPFDOC 
 Price of chicks 
0.61 
LNPLB 
 Wage Rate 
1.64 
LNPF 
 Price of feeds 
1.22 
LNPSH 
 Price of output (Broiler) 
3.91 
LNFLAB 
 Family labour/Q 
n.s. 
LNVK 
 Value of capital stock/Q 
n.s. 
LNPHLB 
 Slope of labour housing 
n.s. 
LN (PLB*FLAB) 
 Wage Rate x Family Labour 
n.s. 
FCR 
 Feed Conversion Ratio 
0.29 
SCALE DUMMY 
 Small vs. Large units 
0.46 
Constt. 

1.49 
Notes: No. of observations (Pooled data of small and large units): 135 Mean Efficiency: 0.683
* We have not taken contract farmers in this analysis of Profit function because the number of observations of contract farmers is small (N=23).
n.s.= not significant
Regression results are based on computer package FRONTIER IV
Source of Data: Indian Poultry Survey, 2002
Table 7.7: Indian Layers: Inefficiency Effects due to Different Variables of Transaction Costs, Policy Distortions, and Environmental Costs Regression Results 

Dependent Variable: Technical Inefficiency (ve = closer to frontier) 

Independent Variables 
Coefficients (d) 
Constant 
4.64 
Age of the decision maker DECAGE 
3.36 
Regional Character (Dummy) REGDUM 
6.80 
Pollution abatement costs ENVTOSTQ 
6.29 
Education of Decision MakerEDUCDM 
n.s. 
Has access to credit, dummy CREDITDUM 
n.s. 
Information source (dummy variable = 1 for radio, TV, newspaper) INFORTVN 
n.s. 
Output market distance 
n.s. 
Other variables 
n.s. 
Relative importance of inefficiency effects 
g is 0.994179 (close to 1) with standard error 0.002130. The efficiency effects are important explainers of profits across farms. 
Notes: No. of observations (Pooled data of small and large units) = 158
n.s.= not significant
Regression results are based on computer package FRONTIER IV
Source of Data: Indian Poultry Survey, 2002
Table 7.8: Indian Broilers: Inefficiency Effects due to Different Variables of Transaction Costs, Policy Distortions, and Environmental Costs Regression Results 

Dependent Variable: Technical Inefficiency (ve = closer to frontier) 

Independent Variable 
Coefficients (d) 
Constant 
28.98 
Age of the decision maker DECAGE 
n.s. 
Information source (dummy variable = 1 for radio, TV, newspaper) INFORTVN 
n.s. 
Regional Character (Dummy) REGDUM 
2.84 
Pollution abatement costs ENVTOSTQ 
n.s. 
Education of Decision MakerEDUCDM 
5.36 
Has access to credit, dummy CREDITDUM 
n.s. 
Output Market Distance  
n.s. 
Other variables 
n.s. 
Relative importance of inefficiency effects 
g is 0.9547 (close to 1), standard error 0.0202. The efficiency effects are important explainers of profits across farms. 
Notes: No. of observations (Pooled data of small and large units) = 135
n.s.= not significant
Regression results are based on computer package FRONTIER IV
Source of Data: Indian Poultry Survey, 2002
Table 7.9 Indian Layers: Differential Inefficiency of Small & Large Farms: Regression Results 

Dependent Variable: Technical Inefficiency (ve = closer to frontier) 

Independent Variable 
Coefficient (d) 

Small 
Large 

Constant 
0.12 
3.56 
Age of the decision maker DECAGE 
n.s. 
3.43 
Information source (dummy variables) INFORTVN 
n.s. 
n.s. 
Regional character (Dummy) REGDUM 
n.s. 
6.38 
Pollution statement costs ENVTOSTQ 
n.s. 
8.36 
Education of the Decision maker EDUCM 
n.s. 
3.07 
Has access to credit CREDITDUM 
n.s. 
n.s. 
Output market distance OUTMARKDIS 
n.s. 
0.88 
No. of observations 
62 
96 
n.s.= not significant
Regression results are based on computer package FRONTIER IV
Source of Data: Indian Poultry Survey, 2002
Table 7.10: Indian Broilers: Differential Inefficiency of Small & Large Farms 

Dependent Variable: Technical Inefficiency (ve = closer to frontier) 

Independent Variable 
Coefficient (d) 

Large 
Small 

Constant 
2.67 
21.4 
Age of the decision maker DECAGE 
n.s. 
n.s. 
Information source (dummy variables) INFORTVN 
1.88 
n.s. 
Regional character (Dummy) REGDOM 
n.s. 
3.59 
Pollution statement costs ENVTOSTQ 
0.32 
n.s. 
Education of Decision Maker  EDUCDM 
n.s. 
6.32 
Has access to credit CREDITDUM 
n.s. 
n.s. 
Output market distance OUTMARKDIS 
n.s. 
n.s. 
No. of observations 
42 
93 
n.s.= not significant
Regression results are based on computer package FRONTIER IV
Source of Data: Indian Poultry Survey, 2002
Table 7.11 Average Profitability of Independent vs. Contract Farms (Rs./bird) 

Category 
Average profitability (Rs/bird) 
Sample size (No.) 
Independent Broiler: Small 
13.130 
93 
Independent Broiler: Large 
10.930 
43 
Independent Broiler: Total 
12.436 
136 
Contract Broiler: Small 
1.034 
17 
Contract Broiler: Large 
3.164 
6 
Contract Broiler: Total 
1.615 
23 
Source of Data: Indian Poultry Survey, 2002
Figure 7.1 Indian Layers: Mean Profit per Unit Across Size Quintiles
Source of Data: Indian Poultry Survey, 2002
Figure 7.2 Indian Layers: Profitability of Small vs. Large
Source of Data: Indian Poultry Survey, 2002
Figure 7.3 Indian Broilers: Average Profitability of Small vs. Large
Source of Data: Indian Poultry Survey, 2002
Figure 7.4 Average Profitability of Indian Broilers: Independent vs. Contract Farmers
Source of Data: Indian Poultry Survey, 2002
^{[63]} Contract farmers were
not included in this analysis because (a) there are very few observations, and
(b) profitability of contract farmers does not depend on prices of inputs
etc. 