Ginkgo Biloba Yaprak Ekstresi'nde Bulunan Bileşikler

The following is a linear regression function useful for cross-section or pooled data to analyze the inverse relationship between farm size and productivity (Carter, 1984):

Yi = +α βX_i (10)

where, is total value of annual crop output per unit of land for household Y_i i and X_i is farm size. We can extend equation (10) by incorporating other relevant variables that would affect productivity:

Equation (11) is estimated using OLS for farm-household level data. We will have two estimations according to equation (1): first, using total value of crops per farm size and second using total value of crops per cultivated area.

We also estimate area expansion effects by considering the ratio between cultivated area and farm size:

is the ratio between cultivated area and farm size, and other variables are as defined above.

We use a quasi-likelihood estimation method within framework of generalized linear models (GLM) for this fractional dependent variable because the predicted values from OLS regression may lie outside the range of 0 and 1, and the conditional variance is not likely to be independent of the conditional mean (Papke and Wooldridge, 1996). The cultivated area share lies between 0 and 1 as only few households have rented-in land and even those rented-in plots were too small to make the cultivated area exceed the total farm size.

For plot level data analysis, equation (11) can be expanded as follows:

1 2 sq 3 h

ip i ip ip i ip

Y = +α β X +β X +β X + +μ ε (12)

where: Y_ip is output value from plot p per unit of land for household i, X_ip^sq is observed plot characteristics, X_ip^h is plot in-variant farm household characteristics, βs are parameters to be estimated, μ_i refers to unobserved plot in-variant household attributes such as farming skills, risk, household time preference, etc. and plot variant attributes (e.g. soil fertility), and ε_ip is the error term.

With the assumption thatμ_i is uncorrelated with , equation (12) is estimated using a household random effects (RE) model. A range of plot level variables designed to capture land quality were included in the RE model for plot level estimation. Household fixed effects could have been used to control for unobserved household and plot characteristics that are plot invariant and to control for intra-group correlation due to unobserved cluster effects (Heltberg, 1998; Udry, 1995; Wooldridge, 2002). However, the variables of interest do not vary over plots within households and would not be captured by the fixed effects model as these important variables will be excluded from estimation during the differencing process.

Using RE enables us to include these variables. We included as many household level variables as possible in our OLS model for the farm household level analysis.

h

Xip

Variable specifications

The dependent variables for the productivity analysis at household level was total output value of crops per total farm size, and decomposed to yield per unit cropped land and area share of cropped land. Only the yield response could be examined with the farm plot level data. Output value was calculated by multiplying crop produce from each plot by average local producer prices. We used the same average prices for both net sellers and buyers of the agricultural outputs because all outputs in the area are traded in the local market and thus we assume low transaction costs in these output markets. Outputs from both main and intercropped (minor) crops were included in the output value from intercropped plots.

Based on our theoretical framework and estimation specifications in equations (10) and (11), we used a number of explanatory variables for the household level farm size-productivity relationship analysis. Female workforce and male workforce refer to family labour in adult-equivalent units per unit of land (timad). We expect both to have positive effects on productivity. For some activities, there is a distinct division of labour between female and

male workforce. The problem of moral hazard is not expected in this specific situation, as nearly all labour is family labour.

Livestock ownership in terms of total livestock units (TLU) may have positive effects on yield as it can supply more manure or can solve cash liquidity problems but may also cause a smaller share of the farm to be under crops if it is set aside for fodder production. Manure and fertilizer applied in each plot have been recorded, and the use of these inputs is expected to improve soil quality and positively affect productivity. However, endogeneity problems may limit the direct use of these variables in the analysis. We included models where we have used the predicted values for fertilizer. It is expected that use of fertilizer will improve soil fertility and thereby productivity of the land. We did not include manure as livestock is included in the regression and may capture both the immediate and lagged effect of manure on productivity.

Endogeneity problems may also be expected in using off-farm income as an explanatory variable, and we could not find good instruments for tackling this problem. We excluded this variable from the productivity regression, as we think all the variables that could affect this endogenous variable are already included in the regression. We were also not able to use crop choice due to endogeneity problems. But we included a dummy for enset presence in one of the plot level models.

We expect farm size, workforce, dependency ratio, and livestock asset to correlate with the cultivated area share, equation (11a). Farm size may have positive relation if small farmers are not intensifying through area expansion. Availability of workforce can facilitate area expansion, and higher dependency ratio may also lead to area expansion in order to fulfill subsistence requirements.

A range of plot level land quality indicators are used in order to control for land quality variations in the plot level analysis that was based on equation (12). Steep slope and shallow depth are expected to negatively affect productivity. Sandy soils may have lower productivity due to low moisture retention. As plot distance increases, households may not be in a position to add farmyard manure. Such plots are therefore likely to have lower productivity.

Conservation structures are present on some of the sloping plots and length of conservation structures was included as a control variable in two of the models in the plot level analysis.

Homestead plots covered by enset plants where much of the manure is applied may be more productive than distant plots covered by annual crops. Enset is expected to help for moisture

retention due to its canopy. Use of fertilizer is expected to boost productivity on the plot. The definition and description of all the variables used for the farm size-land productivity analyses at both levels are shown in Table 3.