Abstract
This paper develops a framework for analysing the sources of total factor productivity (TFP) changes by explicitly taking into account the damage-control nature of pesticides. In the proposed framework, TFP changes are attributed to the conventional sources of growth (i.e. technical change, scale effect and changes in technical efficiency) and the damage-control effect which consists of three distinct components: the first one is due to changes in the initial pest infestation, the second is a spillover effect arising from neighbours' use of preventive inputs and the third is related to abatement effectiveness. The proposed model is applied to a panel of olive-growing farms in Crete, Greece, during the period 1999–2003. The empirical results indicate that the damage-control effect accounted, on average, for 5 per cent of the annual TFP growth and its main component was the improvements in abatement effectiveness.
1. Introduction
The damage-control nature of pesticides has not been considered in any previous study of total factor productivity (TFP). Instead pesticides have been treated as a conventional input that affects output directly, while in reality their contribution is indirect, through their ability to reduce crop damage due to pest infestation and diseases. The way pesticides appear in the production function depends on whether pesticides are treated as a damage-control or an output-expanding input. Specifically, as a damage-control input, pesticides enter into the production function indirectly through either the abatement (Lichtenberg and Zilberman, 1986) or the output damage function (Fox and Weersink, 1995). Consequently, the way of calculating pesticides' marginal product and its output elasticity should be reconsidered. In fact, the results of previous empirical studies (e.g. Lichtenberg and Zilberman, 1986; Oude Lansink and Carpentier, 2001; Oude Lansink and Silva, 2004) indicate that the marginal product of pesticides tends to be overestimated when it is modelled as a conventional rather than a damage-control input.
This bias in the estimated marginal product of pesticides is going, among other things, to affect both the measurement of TFP changes (if output elasticities instead of cost shares are used to compute input growth) and its decomposition, through the magnitude and the relative importance of the scale effect. However, the direction of the impact cannot be predicted with certainty: the upward bias in the estimated marginal product of pesticides results in a greater output elasticity and consequently, in an overestimation of scale elasticity compared with the case of treating pesticides as a damage-control input. On the other hand, it also implies that the contribution of conventional inputs to the growth of aggregate input, defined as a weighted average over all inputs with the ratios of output to scale elasticity used as weights (Chan and Mountain, 1983), would be understated, while that of pesticides would be overstated. Thus, the net effect on the growth of aggregate input cannot be determined a priori. This in turn implies that the impact on the scale effect, which depends on both the magnitude of the scale elasticity and the growth of aggregate input, is ambiguous.
This paper develops a framework for analysing the sources of TFP changes by explicitly taking into account the damage-control nature of pesticides. In the proposed framework, TFP changes are attributed to the conventional sources of growth (namely, technical change, scale effect and changes in technical efficiency) and the damage-control effect, which consists of three distinct components: the first one is due to changes in the initial pest infestation, the second is a spillover effect arising from neighbours' use of preventive inputs and the third is related to abatement effectiveness. To develop this decomposition framework, we extend the output damage approach into two directions: first, we analyse the damage-control nature of pesticides in the presence of technical inefficiency, and second, we introduce a spillover variable into the abatement technology specification.
The rest of this paper is organised as follows: the theoretical model for analysing the damage-control effect of pesticides on TFP changes is presented in the next section. The empirical model and the estimation procedure are discussed in Section 3. The data are presented in Section 4 and the empirical results are analysed in Section 5. Concluding remarks follow in the last section.
2. Theoretical framework
Following Zhengfei et al. (2006), we may distinguish among three production levels depending upon the entailed growth conditions. First, potential output is the highest possible production achievable with given physical environmental and genetic plant characteristics (both beyond farmers' control) assuming agronomic non-limiting use of water and nutrients. On the other hand, shortage or use below their agronomic non-limiting level of water and/or nutrients results in effective or attainable output, which is less than potential output and is achievable assuming the absence of crop damage agents such as weeds, diseases, pests and pollutants. The presence of damage agents may lower further production to the level of actual output. The main damage-control tool in the last decades is the use of pesticides, which take the form of chemical herbicides, fungicides and insecticides. In contrast to other agricultural inputs, pesticides pose no output-expanding features but rather prevent output from undesired and harmful pests' attacks. The damage-control nature of pesticides has led to an asymmetric treatment in relation to other agricultural inputs when modelling production technology.
There are two alternative approaches for incorporating pesticides as a damage-control input into a production function: the abatement and the output damage function approach (Lichtenberg and Zilberman, 1986; Fox and Weersink, 1995). In the former, it is assumed that the true measured impact of pesticides on the effective (attainable) output is related to the purchased abatement rather than the quantity of pesticides used. As a result, abatement rather than pesticides, enters directly into the production function since the former is considered an intermediate input produced by pesticides. In such a setting, the marginal productivity of pesticides reflects their ability to reduce crop damage due to pest infestation and not to increase output directly. In the abatement function approach, it is assumed that the marginal productivity of pesticides is decreasing, which sometimes may be, thought of as a limitation. It is also assumed that the abatement function is independent of initial pest infestation. This implies that the abatement function approach is an appropriate modelling alternative when pesticides are applied in a prophylactic way according to an in-advanced planned schedule.1 If, however, farmers postpone spraying until realisation of pest incidence, the abatement function approach results in biased estimates of the production function parameters (Hall and Moffitt, 2002) because the error term, which necessarily includes the omitted from the abatement function's initial level of pest infestation, is correlated with pesticide use. In such a case, the marginal productivity of pesticides tends to be underestimated (Norwood and Marra, 2003).
On the other hand, in the output damage function approach, it is assumed that the effect of pesticides on the effective (attainable) output is the result of a process involving two stages: (i) the effect of the damage-control input on the damage agent (abatement), and (ii) the effect of the remaining damage agent on the effective output. In the first stage, pest incidence depends on the untreated pest population and on the proportion of it controlled by the abatement activities. In the second stage, effective (attainable) output is indirectly affected by abatement through the loss caused by the remaining damage agent incidence. By construction, the output damage function approach is more appropriate when pesticides are applied once pest incidence is realised and in addition, for certain specifications of the damage-control function, allows for increasing marginal product of pesticides. The case of increasing returns is important from a policy point of view as measures aimed to reduce pesticide use for environmental conservation by imposing a tax may have substantially different effects on the levels of different products.2
For the purposes of this paper, which are mainly related to olive trees, the crop considered in the empirical application of the model, we employ the output damage approach, as spraying for Bactrocera oleae (Gmellin), the main pest in olive tree cultivation, is done once pest incidence is realised. Nevertheless, the framework for the decomposition of TFP developed below is general enough to be used, after making the necessary adjustments, within an abatement function approach. In particular, as will be evident from equation (7), the only difference is that the component of the damage-control effect which is due to changes in initial pest infestation will be absent in the case of the abatement function approach. Thus, in this case, the damage-control effect consists of the spillover effect related to neighbours' use of pesticides and the abatement effectiveness effect.
Following Fox and Weersink (1995), the damage caused in output by pest incidence can be represented by a non-decreasing and concave function
, and
, which measures the proportion of output loss for any given level of pest incidence.3 If the damage agent is absent (b= 0), then
and realised (actual) output equals effective output. If, however, the level of damage agent population tends to infinity
then
and realised output approaches a minimum level
which reflects the maximum destructive capacity of damage agents. On the other hand, pest incidence (density) depends on the initial level of pest population
and the proportion of the damage agent that is not controlled for a given level of treatment (Fox and Weersink, 1995), that is,
, where
is the control function.
In this paper, we enrich Fox and Weersink's (1995) specification of the control function in two ways. In particular, we assume that the proportion of the damage agent remaining after treatment depends not only on the quantity of pesticides used by the individual farmer but also (i) on the total amount of pesticides used by the neighbouring farms and (ii) on abatement effectiveness, which is related to improved field coverage, higher eradication levels, etc., as suggested by Morrison-Paul (1999: 185).4 Thus, the control function is defined as
where
refers to the quantity of pesticides used by each farmer,
is the total amount of pesticides used by the neighbouring farms and t is a time trend reflecting changes in abatement effectiveness. If
pesticides have no effect on damage agent incidence and the level of damage agent affecting farm production is equal to its initial population
. If, however,
there is a complete eradication of the damage agent and actual and effective (attainable) outputs coincide.
The control function is non-decreasing and concave in pesticides use, abatement effectiveness and the spillover variable
.5 The latter implies that there may be some synergies in the use of pesticides (e.g. positive externality). This seems quite reasonable within small geographical areas and particularly for mobile pest populations where the preventive action of every farmer accounts for the total damage caused, and the aggregate intensity of the abatement effort in the area affects the individual damage-control decisions. As the aggregate intensity of abatement effort increases, because more farmers are involved in the use of pesticides or they use them more intensively, lesser doses are required by each farmer to ensure the same level of output damage. Consequently, for a given level of pesticide use, individual abatement does not deteriorate as neighbouring farms increase their pesticide use, and vice versa. In cases, however, that production takes place under controlled or protected conditions (i.e. glasshouses), neighbours' abatement effort does not affect pest incidence and hence the spillover effect is zero.
The second term in equation (7) refers to the scale effect. The sign and direction of this term depend both on the magnitude of the scale elasticity and the over-time changes in the aggregate input, which is given by a Divisia aggregate of conventional and preventive inputs. The scale effect is positive (negative) under increasing (decreasing) returns to scale as long as the aggregate input increases and vice versa. This term vanishes when either technology exhibits constant returns to scale (CRTS) with respect to both conventional and preventive inputs or the aggregate input remains unchanged over time.
The sum of the third and the fourth terms in equation (7) is the technical efficiency changes effect that may be due to either passage of time (e.g. learning-by-doing) (third term) or changes in farm-specific characteristics affecting the managerial and organisational ability of farmers (fourth term).7 They contribute positively (negatively) to TFP growth as long as efficiency changes are associated with movements towards (away from) the production frontier. The technical efficiency change effect is zero and thus has no impact on TFP growth when technical efficiency and all farm-specific characteristics are time invariant (Karagiannis and Tzouvelekas, 2005).
The sum of the last three terms in equation (7), which we refer to it as the damage-control effect, results from treating pesticides as a preventive rather than an output-expanding input. All three components of the damage-control effect contribute to TFP changes through greater actual output rather than through input conservation, reflecting the output-expanding nature of abatement activities. In addition, we should notice that the fifth term in equation (7), which is related to the effect of initial pest infestation to TFP growth, will be absent within an abatement function framework, as in this case pest density does not depend upon initial pest population. Then the damage-control effect consists of the spillover effect related to neighbours' use of pesticides and the abatement effectiveness effect.
The first component of the damage-control effect reflects the impact of the initial pest infestation [fifth term in equation (7)] and, given that , it contributes negatively (positively) to TFP growth as long as initial pest incidence increases (decreases) over time, while it has no impact on TFP growth when
. Since actual output will be lower (greater) with an increase (decrease) in the initial level of pest incidence, this is going to have a negative (positive) productivity effect because less actual output could be obtained from any given amount of conventional and preventive inputs. Thus, unfavourable conditions for pest reproduction, related to the biological cycle of the damage agent, environmental conditions, etc., may enhance TFP growth as fewer pests harm farm output and hence less damage occurs in realised output, and vice versa.
The spillover effect [sixth term in equation (7)] is the second component of the damage-control effect. In the presence of synergies in pesticide use the spillover effect has a positive (negative) impact on TFP growth if the total quantity of preventive inputs used by the neighbouring farms increases (decreases) over time. Since actual output will be greater (lower) with an increase (decrease) in the aggregate abatement effort of neighbouring farms, this is going to have a positive (negative) productivity effect as more actual output could be obtained from any given amount of conventional and preventive inputs. The spillover effect has no impact on TFP growth if either production takes place under controlled or protected conditions (i.e. glasshouses) and thus neighbours' abatement effort does not affect production
or aggregate abatement effort of neighbours remains unchanged over time
8
The last component of the damage-control effect [seventh term in equation (7)] is related to the rate of abatement effectiveness. Since effective (attainable) output will be greater with an improvement in abatement effectiveness, this is going to have a positive productivity effect as more effective (attainable) and more actual output could be obtained from any given amount of conventional and preventive inputs. However, the rate of abatement effectiveness does not contribute in a one-to-one relationship to TFP growth as does technical change, but its contribution is proportional to the product of the marginal damage effect and the ratio of the uncontrolled pest to the percentage of output loss
. Abatement effectiveness does not contribute to TFP growth if either the initial pest infestation is equal to realised pest incidence, i.e.
, or the rate of abatement effectiveness remains unchanged, i.e.
. In the former case, the marginal effectiveness of abatement activities is zero, which means that abatement has no effect on initial pest infestation. This would be the case with incorrect application or inappropriate choice of pesticides for both the farmer and his/her neighbours
Lastly, it is worth mentioning that if someone is interested in the measurement of the overall impact of pesticides on TFP growth, then the contribution of pesticides through the scale effect, i.e. , should be added to the damage-control effect. For this reason, this term is presented separately in the decomposition analysis (see Table 6).
3. Empirical model
We examine the robustness of the proposed specification using the procedure suggested by Zhengfei et al. (2005). Specifically, pesticides are also included into in equation (12) to create an augmented model. Then, two hypotheses can be tested: first, the coefficient of pesticides in
is 0 and second, the value of
is equal to 1. If the first hypothesis is rejected whereas the second is not, the data favour the treatment of pesticides as an output-expanding input, symmetrically to all other agricultural inputs. In contrast, if the first hypothesis is not rejected whereas the second is rejected, the data favour the asymmetric treatment of pesticides as a damage-control input.
4. Data
The data for this study are taken from the Greek National Agricultural Research Foundation (NAgReF) and refer to 60 olive-growing farms in the island of Crete, during the period 1999–2003, namely a balanced panel data set. The data set contain information on production volumes and input expenses as well as pesticides application against olive fruit fly Bactrocera oleae (Gmellin). Farms were located in the same geographical area in the western part of the island and were specialised in olive tree cultivation. In addition, the data set contains several farm-specific information including demographic characteristics, environmental conditions and extension services provision.
One output and four inputs are considered in the empirical model. Output is given by total olive oil production measured in kilograms. The inputs considered are labour (including family and hired workers) measured in hours, land measured in stremmas (one stremma equals 0.1 ha) and, other costs, consisting of fuel and electric power, fertilisers, storage, and irrigation water expenditures, measured in euros.11 All variables measured in money terms have been converted into 2003 constant values.
On the other hand, pest population was measured using chemical traps installed in every 500 m2 of farm's plots. The number of flies captured in the traps was used to extrapolate the whole pest population in each plot. The damage abatement input includes pesticide materials measured in litres. The pesticide spillover variable was constructed by aggregating pesticide use of all neighbouring farms. Following suggestions by the local Agricultural Experimental Stations' agronomists, we define neighbouring farms as those located in the same small area with similar micro-climatic conditions that affect pest population and its biological cycle.
In the inefficiency effect function, we include the following variables which are assumed to affect efficiency differentials: farm owner's education, measured in years of schooling, the family size measured as the number of persons in the household, an aridity index defined as the ratio of the average annual temperature in the region over the total annual precipitation (Stallings, 1960), the altitude of farms' location measured in metres and the number of extension visits in the farm.
In order to avoid problems associated with units of measurement, all variables included in equation (13) were normalised. The basis for normalisation was the farm with the smallest deviation of its output and input levels from the sample means. Summary statistics of the variables used in the empirical model are given in Table 1.
Summary statistics of the variables in the empirical model
| Variable | Average | Minimum | Maximum | Standard deviation |
|---|---|---|---|---|
| Output (kg) | 18,758.7 | 2,049.7 | 71,789 | 13924.5 |
| Land (ha) | 0.87 | 0.12 | 6.6 | 0.75 |
| Labour (h) | 2,867.0 | 437.0 | 12,320 | 1912.6 |
| Other cost (EUR) | 3,407.8 | 74.3 | 22,249 | 2298.0 |
| Pesticides (l) | 1,833.7 | 71.0 | 9,604 | 1493.7 |
| Spillover (l) | 5,487.0 | 974.0 | 16,339 | 3023.3 |
| Family size (number of persons) | 4.0 | 1.0 | 9 | 1.4 |
| Altitude (m) | 276.2 | 1.0 | 995 | 270.6 |
| Aridity index | 1.03 | 0.03 | 3.52 | 0.78 |
| Extension visits (number of visits) | 5.8 | 0.0 | 26 | 5.2 |
| Pest population (pests per m2) | 15.7 | 0.1 | 54.2 | 14.6 |
| Education (years) | 7.1 | 1.0 | 16 | 3.8 |
| Variable | Average | Minimum | Maximum | Standard deviation |
|---|---|---|---|---|
| Output (kg) | 18,758.7 | 2,049.7 | 71,789 | 13924.5 |
| Land (ha) | 0.87 | 0.12 | 6.6 | 0.75 |
| Labour (h) | 2,867.0 | 437.0 | 12,320 | 1912.6 |
| Other cost (EUR) | 3,407.8 | 74.3 | 22,249 | 2298.0 |
| Pesticides (l) | 1,833.7 | 71.0 | 9,604 | 1493.7 |
| Spillover (l) | 5,487.0 | 974.0 | 16,339 | 3023.3 |
| Family size (number of persons) | 4.0 | 1.0 | 9 | 1.4 |
| Altitude (m) | 276.2 | 1.0 | 995 | 270.6 |
| Aridity index | 1.03 | 0.03 | 3.52 | 0.78 |
| Extension visits (number of visits) | 5.8 | 0.0 | 26 | 5.2 |
| Pest population (pests per m2) | 15.7 | 0.1 | 54.2 | 14.6 |
| Education (years) | 7.1 | 1.0 | 16 | 3.8 |
Summary statistics of the variables in the empirical model
| Variable | Average | Minimum | Maximum | Standard deviation |
|---|---|---|---|---|
| Output (kg) | 18,758.7 | 2,049.7 | 71,789 | 13924.5 |
| Land (ha) | 0.87 | 0.12 | 6.6 | 0.75 |
| Labour (h) | 2,867.0 | 437.0 | 12,320 | 1912.6 |
| Other cost (EUR) | 3,407.8 | 74.3 | 22,249 | 2298.0 |
| Pesticides (l) | 1,833.7 | 71.0 | 9,604 | 1493.7 |
| Spillover (l) | 5,487.0 | 974.0 | 16,339 | 3023.3 |
| Family size (number of persons) | 4.0 | 1.0 | 9 | 1.4 |
| Altitude (m) | 276.2 | 1.0 | 995 | 270.6 |
| Aridity index | 1.03 | 0.03 | 3.52 | 0.78 |
| Extension visits (number of visits) | 5.8 | 0.0 | 26 | 5.2 |
| Pest population (pests per m2) | 15.7 | 0.1 | 54.2 | 14.6 |
| Education (years) | 7.1 | 1.0 | 16 | 3.8 |
| Variable | Average | Minimum | Maximum | Standard deviation |
|---|---|---|---|---|
| Output (kg) | 18,758.7 | 2,049.7 | 71,789 | 13924.5 |
| Land (ha) | 0.87 | 0.12 | 6.6 | 0.75 |
| Labour (h) | 2,867.0 | 437.0 | 12,320 | 1912.6 |
| Other cost (EUR) | 3,407.8 | 74.3 | 22,249 | 2298.0 |
| Pesticides (l) | 1,833.7 | 71.0 | 9,604 | 1493.7 |
| Spillover (l) | 5,487.0 | 974.0 | 16,339 | 3023.3 |
| Family size (number of persons) | 4.0 | 1.0 | 9 | 1.4 |
| Altitude (m) | 276.2 | 1.0 | 995 | 270.6 |
| Aridity index | 1.03 | 0.03 | 3.52 | 0.78 |
| Extension visits (number of visits) | 5.8 | 0.0 | 26 | 5.2 |
| Pest population (pests per m2) | 15.7 | 0.1 | 54.2 | 14.6 |
| Education (years) | 7.1 | 1.0 | 16 | 3.8 |
5. Empirical results
The robustness of the proposed specification [equations (13) and (14)] was examined using the testing procedure suggested by Zhengfei et al. (2005). The results of this statistical test favour the asymmetric treatment of pesticide as a damage-control rather than an output-expanding input. In particular, the test that the parameters associated with pesticides in are statistically significant yielded a p-value of 0.342, whereas that for
being equal to 1 was 0.003.12 The ML parameter estimates of the supported model are presented in Table 2. All the estimated parameters of the conventional inputs and of pesticides have the anticipated magnitude and sign and the majority of them are statistically significant at the 5 per cent level. As a result, concavity of the production function with respect to both conventional and preventive inputs is satisfied at the point of normalisation. This means that the marginal product of both conventional and damage-control inputs is positive and diminishing. Thus, although our model specification allows for the presence of increasing marginal returns to pesticides, the data do not support that finding. Average values of the estimated output elasticities and marginal products are given in Table 3.
Parameter estimates of the generalised Cobb–Douglas stochastic production frontier
| Variable | Estimate | Standard error |
|---|---|---|
| Production frontier | ||
| β0 | −0.0313 | 0.0751 |
| βA | 0.3355** | 0.0401 |
| βL | 0.3109** | 0.0408 |
| βC | 0.2134** | 0.0355 |
| βT | 0.2067* | 0.1032 |
| βTT | −0.2026 | 0.1237 |
| βTA | 0.0552 | 0.0673 |
| βTL | −0.1259* | 0.0704 |
| βTC | −0.0799* | 0.0521 |
| λ | −0.0410* | 0.0175 |
| ζP | 0.8602** | 0.2135 |
| ζS | 0.0085* | 0.0039 |
| ζT | 0.0148** | 0.0052 |
| Inefficiency effects model | ||
| δ0 | −0.1749 | 0.7111 |
| δFAM | −0.5433* | 0.2447 |
| δALT | 0.0027 | 0.0757 |
| δARD | 0.7554* | 0.3482 |
| δEXT | −0.5522* | 0.2368 |
| δEDU | −0.3602** | 0.1362 |
| δT | −0.2258** | 0.1087 |
| δTT | 0.3459 | 0.2855 |
| γ | 0.9105** | 0.0507 |
| σ2 | 0.6368** | 0.2366 |
| Variable | Estimate | Standard error |
|---|---|---|
| Production frontier | ||
| β0 | −0.0313 | 0.0751 |
| βA | 0.3355** | 0.0401 |
| βL | 0.3109** | 0.0408 |
| βC | 0.2134** | 0.0355 |
| βT | 0.2067* | 0.1032 |
| βTT | −0.2026 | 0.1237 |
| βTA | 0.0552 | 0.0673 |
| βTL | −0.1259* | 0.0704 |
| βTC | −0.0799* | 0.0521 |
| λ | −0.0410* | 0.0175 |
| ζP | 0.8602** | 0.2135 |
| ζS | 0.0085* | 0.0039 |
| ζT | 0.0148** | 0.0052 |
| Inefficiency effects model | ||
| δ0 | −0.1749 | 0.7111 |
| δFAM | −0.5433* | 0.2447 |
| δALT | 0.0027 | 0.0757 |
| δARD | 0.7554* | 0.3482 |
| δEXT | −0.5522* | 0.2368 |
| δEDU | −0.3602** | 0.1362 |
| δT | −0.2258** | 0.1087 |
| δTT | 0.3459 | 0.2855 |
| γ | 0.9105** | 0.0507 |
| σ2 | 0.6368** | 0.2366 |
Note: A stands for land; L, labour; C, other cost; T, time; FAM, family size; P, pesticides; S, pesticides spillovers; ALT, altitude; ARD, the aridity index; EXT, the number of extension visits; EDU, farmer's education level.
*Statistical significance at the 5 per cent level.
**Statistical significance at the 1 per cent level.
Parameter estimates of the generalised Cobb–Douglas stochastic production frontier
| Variable | Estimate | Standard error |
|---|---|---|
| Production frontier | ||
| β0 | −0.0313 | 0.0751 |
| βA | 0.3355** | 0.0401 |
| βL | 0.3109** | 0.0408 |
| βC | 0.2134** | 0.0355 |
| βT | 0.2067* | 0.1032 |
| βTT | −0.2026 | 0.1237 |
| βTA | 0.0552 | 0.0673 |
| βTL | −0.1259* | 0.0704 |
| βTC | −0.0799* | 0.0521 |
| λ | −0.0410* | 0.0175 |
| ζP | 0.8602** | 0.2135 |
| ζS | 0.0085* | 0.0039 |
| ζT | 0.0148** | 0.0052 |
| Inefficiency effects model | ||
| δ0 | −0.1749 | 0.7111 |
| δFAM | −0.5433* | 0.2447 |
| δALT | 0.0027 | 0.0757 |
| δARD | 0.7554* | 0.3482 |
| δEXT | −0.5522* | 0.2368 |
| δEDU | −0.3602** | 0.1362 |
| δT | −0.2258** | 0.1087 |
| δTT | 0.3459 | 0.2855 |
| γ | 0.9105** | 0.0507 |
| σ2 | 0.6368** | 0.2366 |
| Variable | Estimate | Standard error |
|---|---|---|
| Production frontier | ||
| β0 | −0.0313 | 0.0751 |
| βA | 0.3355** | 0.0401 |
| βL | 0.3109** | 0.0408 |
| βC | 0.2134** | 0.0355 |
| βT | 0.2067* | 0.1032 |
| βTT | −0.2026 | 0.1237 |
| βTA | 0.0552 | 0.0673 |
| βTL | −0.1259* | 0.0704 |
| βTC | −0.0799* | 0.0521 |
| λ | −0.0410* | 0.0175 |
| ζP | 0.8602** | 0.2135 |
| ζS | 0.0085* | 0.0039 |
| ζT | 0.0148** | 0.0052 |
| Inefficiency effects model | ||
| δ0 | −0.1749 | 0.7111 |
| δFAM | −0.5433* | 0.2447 |
| δALT | 0.0027 | 0.0757 |
| δARD | 0.7554* | 0.3482 |
| δEXT | −0.5522* | 0.2368 |
| δEDU | −0.3602** | 0.1362 |
| δT | −0.2258** | 0.1087 |
| δTT | 0.3459 | 0.2855 |
| γ | 0.9105** | 0.0507 |
| σ2 | 0.6368** | 0.2366 |
Note: A stands for land; L, labour; C, other cost; T, time; FAM, family size; P, pesticides; S, pesticides spillovers; ALT, altitude; ARD, the aridity index; EXT, the number of extension visits; EDU, farmer's education level.
*Statistical significance at the 5 per cent level.
**Statistical significance at the 1 per cent level.
Output elasticities, returns to scale, damage abatement elasticities and marginal products of conventional inputs and pesticides
| 1999 | 2000 | 2001 | 2002 | 2003 | Mean | |
|---|---|---|---|---|---|---|
| Output elasticities | ||||||
| Land | 0.4738** | 0.3866** | 0.3355** | 0.2993** | 0.2712** | 0.3533** |
| Labour | 0.4492** | 0.3619** | 0.3109** | 0.2746** | 0.2465** | 0.3286** |
| Other cost | 0.3011** | 0.2458** | 0.2134** | 0.1904** | 0.1726** | 0.2247** |
| Pesticides | 0.0186** | 0.0115** | 0.0053** | 0.0176** | 0.0225** | 0.0151** |
| Returns to scale | 1.2428** | 1.0057** | 0.8650** | 0.7819** | 0.7128** | 0.9217** |
| Damage-control elasticities | ||||||
| Initial pest population | −0.0108* | −0.0067* | −0.0031* | −0.0102* | −0.0131* | −0.0088* |
| Pesticides spillover | 0.0018* | 0.0011* | 0.0005* | 0.0017* | 0.0022* | 0.0015* |
| Marginal products | ||||||
| Land (EUR/ha) | 1,357 | 1,048 | 659 | 949 | 559 | 914 |
| Labour (EUR/h) | 3.4873 | 3.9777 | 3.4174 | 1.5160 | 1.3191 | 2.7435 |
| Other cost (EUR/ EUR) | 2.0294 | 1.8410 | 1.3889 | 1.0669 | 1.4406 | 1.5534 |
| Pesticides (EUR/l) | 0.2206 | 0.1097 | 0.1654 | 0.2776 | 0.2591 | 0.2065 |
| 1999 | 2000 | 2001 | 2002 | 2003 | Mean | |
|---|---|---|---|---|---|---|
| Output elasticities | ||||||
| Land | 0.4738** | 0.3866** | 0.3355** | 0.2993** | 0.2712** | 0.3533** |
| Labour | 0.4492** | 0.3619** | 0.3109** | 0.2746** | 0.2465** | 0.3286** |
| Other cost | 0.3011** | 0.2458** | 0.2134** | 0.1904** | 0.1726** | 0.2247** |
| Pesticides | 0.0186** | 0.0115** | 0.0053** | 0.0176** | 0.0225** | 0.0151** |
| Returns to scale | 1.2428** | 1.0057** | 0.8650** | 0.7819** | 0.7128** | 0.9217** |
| Damage-control elasticities | ||||||
| Initial pest population | −0.0108* | −0.0067* | −0.0031* | −0.0102* | −0.0131* | −0.0088* |
| Pesticides spillover | 0.0018* | 0.0011* | 0.0005* | 0.0017* | 0.0022* | 0.0015* |
| Marginal products | ||||||
| Land (EUR/ha) | 1,357 | 1,048 | 659 | 949 | 559 | 914 |
| Labour (EUR/h) | 3.4873 | 3.9777 | 3.4174 | 1.5160 | 1.3191 | 2.7435 |
| Other cost (EUR/ EUR) | 2.0294 | 1.8410 | 1.3889 | 1.0669 | 1.4406 | 1.5534 |
| Pesticides (EUR/l) | 0.2206 | 0.1097 | 0.1654 | 0.2776 | 0.2591 | 0.2065 |
Note: Standard errors were obtained using block re-sampling techniques which entails grouping the data randomly in a number of blocks of farms and re-estimating the model leaving out each time one of the blocks of observations (i.e. five) and then computing the corresponding standard errors (Politis and Romano, 1994).
*Statistical significance at the 5 per cent level.
**Statistical significance at the 1 per cent level.
Output elasticities, returns to scale, damage abatement elasticities and marginal products of conventional inputs and pesticides
| 1999 | 2000 | 2001 | 2002 | 2003 | Mean | |
|---|---|---|---|---|---|---|
| Output elasticities | ||||||
| Land | 0.4738** | 0.3866** | 0.3355** | 0.2993** | 0.2712** | 0.3533** |
| Labour | 0.4492** | 0.3619** | 0.3109** | 0.2746** | 0.2465** | 0.3286** |
| Other cost | 0.3011** | 0.2458** | 0.2134** | 0.1904** | 0.1726** | 0.2247** |
| Pesticides | 0.0186** | 0.0115** | 0.0053** | 0.0176** | 0.0225** | 0.0151** |
| Returns to scale | 1.2428** | 1.0057** | 0.8650** | 0.7819** | 0.7128** | 0.9217** |
| Damage-control elasticities | ||||||
| Initial pest population | −0.0108* | −0.0067* | −0.0031* | −0.0102* | −0.0131* | −0.0088* |
| Pesticides spillover | 0.0018* | 0.0011* | 0.0005* | 0.0017* | 0.0022* | 0.0015* |
| Marginal products | ||||||
| Land (EUR/ha) | 1,357 | 1,048 | 659 | 949 | 559 | 914 |
| Labour (EUR/h) | 3.4873 | 3.9777 | 3.4174 | 1.5160 | 1.3191 | 2.7435 |
| Other cost (EUR/ EUR) | 2.0294 | 1.8410 | 1.3889 | 1.0669 | 1.4406 | 1.5534 |
| Pesticides (EUR/l) | 0.2206 | 0.1097 | 0.1654 | 0.2776 | 0.2591 | 0.2065 |
| 1999 | 2000 | 2001 | 2002 | 2003 | Mean | |
|---|---|---|---|---|---|---|
| Output elasticities | ||||||
| Land | 0.4738** | 0.3866** | 0.3355** | 0.2993** | 0.2712** | 0.3533** |
| Labour | 0.4492** | 0.3619** | 0.3109** | 0.2746** | 0.2465** | 0.3286** |
| Other cost | 0.3011** | 0.2458** | 0.2134** | 0.1904** | 0.1726** | 0.2247** |
| Pesticides | 0.0186** | 0.0115** | 0.0053** | 0.0176** | 0.0225** | 0.0151** |
| Returns to scale | 1.2428** | 1.0057** | 0.8650** | 0.7819** | 0.7128** | 0.9217** |
| Damage-control elasticities | ||||||
| Initial pest population | −0.0108* | −0.0067* | −0.0031* | −0.0102* | −0.0131* | −0.0088* |
| Pesticides spillover | 0.0018* | 0.0011* | 0.0005* | 0.0017* | 0.0022* | 0.0015* |
| Marginal products | ||||||
| Land (EUR/ha) | 1,357 | 1,048 | 659 | 949 | 559 | 914 |
| Labour (EUR/h) | 3.4873 | 3.9777 | 3.4174 | 1.5160 | 1.3191 | 2.7435 |
| Other cost (EUR/ EUR) | 2.0294 | 1.8410 | 1.3889 | 1.0669 | 1.4406 | 1.5534 |
| Pesticides (EUR/l) | 0.2206 | 0.1097 | 0.1654 | 0.2776 | 0.2591 | 0.2065 |
Note: Standard errors were obtained using block re-sampling techniques which entails grouping the data randomly in a number of blocks of farms and re-estimating the model leaving out each time one of the blocks of observations (i.e. five) and then computing the corresponding standard errors (Politis and Romano, 1994).
*Statistical significance at the 5 per cent level.
**Statistical significance at the 1 per cent level.
The ratio parameter, , is positive and statistically significant at the 1 per cent level, indicating that the technical inefficiency is likely to have an important effect in explaining output variability among farms in the sample. According to the estimated variances, output variability is mainly due to technical inefficiency rather than statistical noise. We further examined this finding using conventional likelihood ratio test, and the results are presented in the upper panel of Table 4.13 First, the null hypothesis that
is rejected at the 5 per cent level of significance, indicating that the technical inefficiency effects are in fact stochastic and present in the model. Moreover, Schmidt and Lin's (1984) test for the skewness of the composed error term also confirms the existence of technical inefficiency.14 Second, the hypotheses that there are no autonomous changes in technical inefficiency over time
, or that individual characteristics do not affect technical inefficiency
, are also rejected at the 5 per cent level of significance. This is also true when both hypotheses are examined jointly
Mean technical efficiency is found to be 74.76 per cent during the period 1999–2003, implying that output could have been increased substantially if farmers' performance was improved (Table 5). Specifically, a 25.24 per cent increase in olive oil production could have been achieved during this period, without altering total input use. Around 50 per cent of the farms in the sample achieved scores of technical efficiency between 70 and 80 per cent, and the portion of farms with technical efficiency scores below 70 per cent consistently decreased over time. With the exception of 2001, technical efficiency ratings followed an increasing pattern over time as mean efficiency scores raised from 72.5 per cent in 1999 to 77.8 per cent in 2003. This implies that technical efficiency changes have positively contributed to TFP growth.
Model specification tests
| Hypothesis | LR statistic | Critical value (α = 0.05) |
|---|---|---|
| Technical efficiency | ||
| Technical efficiency | 35.14 | |
| No autonomous changes in efficiency | 21.48 | |
| No-individual farm effects | 28.41 | |
| Time invariant efficiency | 36.87 | |
| Structure of production | ||
| CRTS | 41.03 | |
| No technical change | 40.23 | |
| Hicks neutral technical change | 24.15 | |
| Abatement activities | ||
| Zero spillover effects | 4.97 | |
| Unchanged abatement effectiveness | 5.02 | |
| Zero marginal effectiveness of abatement activities | 38.54 | |
| Hypothesis | LR statistic | Critical value (α = 0.05) |
|---|---|---|
| Technical efficiency | ||
| Technical efficiency | 35.14 | |
| No autonomous changes in efficiency | 21.48 | |
| No-individual farm effects | 28.41 | |
| Time invariant efficiency | 36.87 | |
| Structure of production | ||
| CRTS | 41.03 | |
| No technical change | 40.23 | |
| Hicks neutral technical change | 24.15 | |
| Abatement activities | ||
| Zero spillover effects | 4.97 | |
| Unchanged abatement effectiveness | 5.02 | |
| Zero marginal effectiveness of abatement activities | 38.54 | |
aIn this case, the asymptotic distribution of the LR ratio test is a mixed χ2 and the appropriate critical values are obtained from Kodde and Palm (1986).
Model specification tests
| Hypothesis | LR statistic | Critical value (α = 0.05) |
|---|---|---|
| Technical efficiency | ||
| Technical efficiency | 35.14 | |
| No autonomous changes in efficiency | 21.48 | |
| No-individual farm effects | 28.41 | |
| Time invariant efficiency | 36.87 | |
| Structure of production | ||
| CRTS | 41.03 | |
| No technical change | 40.23 | |
| Hicks neutral technical change | 24.15 | |
| Abatement activities | ||
| Zero spillover effects | 4.97 | |
| Unchanged abatement effectiveness | 5.02 | |
| Zero marginal effectiveness of abatement activities | 38.54 | |
| Hypothesis | LR statistic | Critical value (α = 0.05) |
|---|---|---|
| Technical efficiency | ||
| Technical efficiency | 35.14 | |
| No autonomous changes in efficiency | 21.48 | |
| No-individual farm effects | 28.41 | |
| Time invariant efficiency | 36.87 | |
| Structure of production | ||
| CRTS | 41.03 | |
| No technical change | 40.23 | |
| Hicks neutral technical change | 24.15 | |
| Abatement activities | ||
| Zero spillover effects | 4.97 | |
| Unchanged abatement effectiveness | 5.02 | |
| Zero marginal effectiveness of abatement activities | 38.54 | |
aIn this case, the asymptotic distribution of the LR ratio test is a mixed χ2 and the appropriate critical values are obtained from Kodde and Palm (1986).
Frequency distribution of technical efficiency ratings of olive-growing farms in Greece, 1999–2003
| TE (per cent) | 1999 | 2000 | 2001 | 2002 | 2003 | 1999–2003 |
|---|---|---|---|---|---|---|
| 0–10 | 0 | 0 | 0 | 0 | 0 | 0 |
| 10–20 | 0 | 0 | 1 | 0 | 0 | 0 |
| 20–30 | 0 | 2 | 1 | 0 | 0 | 0 |
| 30–40 | 4 | 2 | 7 | 2 | 2 | 0 |
| 40–50 | 1 | 1 | 0 | 4 | 2 | 0 |
| 50–60 | 6 | 6 | 1 | 1 | 5 | 2 |
| 60–70 | 12 | 6 | 9 | 8 | 3 | 13 |
| 70–80 | 12 | 9 | 16 | 12 | 12 | 31 |
| 80–90 | 20 | 30 | 20 | 27 | 31 | 14 |
| 90–100 | 5 | 4 | 5 | 6 | 5 | 0 |
| Mean | 72.46 | 74.81 | 71.69 | 76.99 | 77.82 | 74.76 |
| Minimum | 30.97 | 22.12 | 17.98 | 34.44 | 33.60 | 53.69 |
| Maximum | 93.84 | 94.12 | 92.00 | 93.77 | 94.24 | 86.57 |
| TE (per cent) | 1999 | 2000 | 2001 | 2002 | 2003 | 1999–2003 |
|---|---|---|---|---|---|---|
| 0–10 | 0 | 0 | 0 | 0 | 0 | 0 |
| 10–20 | 0 | 0 | 1 | 0 | 0 | 0 |
| 20–30 | 0 | 2 | 1 | 0 | 0 | 0 |
| 30–40 | 4 | 2 | 7 | 2 | 2 | 0 |
| 40–50 | 1 | 1 | 0 | 4 | 2 | 0 |
| 50–60 | 6 | 6 | 1 | 1 | 5 | 2 |
| 60–70 | 12 | 6 | 9 | 8 | 3 | 13 |
| 70–80 | 12 | 9 | 16 | 12 | 12 | 31 |
| 80–90 | 20 | 30 | 20 | 27 | 31 | 14 |
| 90–100 | 5 | 4 | 5 | 6 | 5 | 0 |
| Mean | 72.46 | 74.81 | 71.69 | 76.99 | 77.82 | 74.76 |
| Minimum | 30.97 | 22.12 | 17.98 | 34.44 | 33.60 | 53.69 |
| Maximum | 93.84 | 94.12 | 92.00 | 93.77 | 94.24 | 86.57 |
Frequency distribution of technical efficiency ratings of olive-growing farms in Greece, 1999–2003
| TE (per cent) | 1999 | 2000 | 2001 | 2002 | 2003 | 1999–2003 |
|---|---|---|---|---|---|---|
| 0–10 | 0 | 0 | 0 | 0 | 0 | 0 |
| 10–20 | 0 | 0 | 1 | 0 | 0 | 0 |
| 20–30 | 0 | 2 | 1 | 0 | 0 | 0 |
| 30–40 | 4 | 2 | 7 | 2 | 2 | 0 |
| 40–50 | 1 | 1 | 0 | 4 | 2 | 0 |
| 50–60 | 6 | 6 | 1 | 1 | 5 | 2 |
| 60–70 | 12 | 6 | 9 | 8 | 3 | 13 |
| 70–80 | 12 | 9 | 16 | 12 | 12 | 31 |
| 80–90 | 20 | 30 | 20 | 27 | 31 | 14 |
| 90–100 | 5 | 4 | 5 | 6 | 5 | 0 |
| Mean | 72.46 | 74.81 | 71.69 | 76.99 | 77.82 | 74.76 |
| Minimum | 30.97 | 22.12 | 17.98 | 34.44 | 33.60 | 53.69 |
| Maximum | 93.84 | 94.12 | 92.00 | 93.77 | 94.24 | 86.57 |
| TE (per cent) | 1999 | 2000 | 2001 | 2002 | 2003 | 1999–2003 |
|---|---|---|---|---|---|---|
| 0–10 | 0 | 0 | 0 | 0 | 0 | 0 |
| 10–20 | 0 | 0 | 1 | 0 | 0 | 0 |
| 20–30 | 0 | 2 | 1 | 0 | 0 | 0 |
| 30–40 | 4 | 2 | 7 | 2 | 2 | 0 |
| 40–50 | 1 | 1 | 0 | 4 | 2 | 0 |
| 50–60 | 6 | 6 | 1 | 1 | 5 | 2 |
| 60–70 | 12 | 6 | 9 | 8 | 3 | 13 |
| 70–80 | 12 | 9 | 16 | 12 | 12 | 31 |
| 80–90 | 20 | 30 | 20 | 27 | 31 | 14 |
| 90–100 | 5 | 4 | 5 | 6 | 5 | 0 |
| Mean | 72.46 | 74.81 | 71.69 | 76.99 | 77.82 | 74.76 |
| Minimum | 30.97 | 22.12 | 17.98 | 34.44 | 33.60 | 53.69 |
| Maximum | 93.84 | 94.12 | 92.00 | 93.77 | 94.24 | 86.57 |
With the exception of altitude, the other farm-specific characteristics considered have had a statistically significant effect on technical efficiency. In particular, it is found that education leads to better utilisation of given inputs as it enables farmers to use technical information more efficiently. Extension services seem also to improve farmers' managerial ability and to affect the efficient utilisation of existing technologies by improving their know-how (Birkhaeuser et al., 1991). Family size tends to result in higher efficiency due to stronger incentives by rural household members. On the other hand, adverse environmental conditions as proxied by the aridity index seem to negatively affect individual efficiency levels.
The next set of hypotheses testing concerns returns to scale and technical change (see the middle panel of Table 4). In particular, the hypothesis of CRTS is rejected at the 5 per cent level of significance. For the whole period under consideration, returns to scale were found to be decreasing. There may be two reasons underlying this finding: either a ‘safety first’ consideration for obtaining a subsistence level of family income and/or the presence of production-based subsidies may have encouraged farmers to operate at a supra-optimal scale. In any case, the scale effect is present and constitutes a source of TFP growth. On the other hand, the hypothesis of no technical change
and that of the Hicks neutral technical change
are both rejected at the 5 per cent significance level and thus technical change has also been an important source of TFP. The average annual rate of technical change is estimated at 1.21 per cent. Regarding technological biases, technical change is found to be labour- and other costs-saving and land-neutral as the relevant estimated parameter was found to be statistically insignificant.
The estimated parameters of the damage and control functions, reported in Table 2, have the anticipated signs and are statistically significant. Based on these, the hypothesis of zero marginal abatement effectiveness is rejected at the 5 per cent level of significance (Table 4). On the other hand, both the hypotheses of a zero-spillover effect and of unchanged abatement effectiveness
are also rejected (Table 4). Thus, pesticides had a positive contribution to damage abatement, with the impact from own use to be much greater than that of the neighbouring farms (see Table 3 for the average estimated values of these impacts in elasticity form). The positive estimated parameter of the spillover variable indicates, however, some (though small) synergies in pesticide use. In addition, there are evidence of improvements in abatement effectiveness, as the relevant estimated parameter (
) is found to be positive and statistically significant. All these advocate the presence of the damage-control effect and its potential role in TFP changes.
The empirical results concerning the decomposition of TFP changes based on equation (7) are reported in Table 6. The average annual rate of TFP growth is estimated at 1.58 per cent. The vast portion (95 per cent) of TFP changes is attributed to the conventional sources of growth (namely, technical change, scale effect and technical efficiency changes) and the remaining 5 per cent to the damage-control effect. Even though the damage-control effect is relatively small compared to the conventional sources of growth, it cannot be neglected by any means as the aforementioned hypotheses testing indicated.
Decomposition of TFP growth of olive-growing farms in Greece, 1999–2003
| Absolute value | Percentage | |
|---|---|---|
| TFP growth | 1.5777 | 100.0 |
| Technical change | 1.2109 | 76.7 |
| Neutral | 0.9413 | 59.7 |
| Biased | 0.2696 | 17.1 |
| Scale effect | −0.1988 | −12.6 |
| Conventional inputs | −0.2060 | −13.1 |
| Pesticides | 0.0072 | 0.5 |
| Technical efficiency change effect | 0.4865 | 30.8 |
| Autonomous changes | 0.2330 | 14.8 |
| Aridity index | 0.1308 | 8.3 |
| Extension services | 0.1227 | 7.8 |
| Damage-control effect | 0.0791 | 5.0 |
| Abatement effectiveness effect | 0.0839 | 5.3 |
| Initial pest infestation effect | −0.0045 | −0.3 |
| Spillover effect | −0.0003 | −0.0 |
| Absolute value | Percentage | |
|---|---|---|
| TFP growth | 1.5777 | 100.0 |
| Technical change | 1.2109 | 76.7 |
| Neutral | 0.9413 | 59.7 |
| Biased | 0.2696 | 17.1 |
| Scale effect | −0.1988 | −12.6 |
| Conventional inputs | −0.2060 | −13.1 |
| Pesticides | 0.0072 | 0.5 |
| Technical efficiency change effect | 0.4865 | 30.8 |
| Autonomous changes | 0.2330 | 14.8 |
| Aridity index | 0.1308 | 8.3 |
| Extension services | 0.1227 | 7.8 |
| Damage-control effect | 0.0791 | 5.0 |
| Abatement effectiveness effect | 0.0839 | 5.3 |
| Initial pest infestation effect | −0.0045 | −0.3 |
| Spillover effect | −0.0003 | −0.0 |
Note: The time invariant farm-specific characteristics are not taken into account in the technical efficiency change effect.
Decomposition of TFP growth of olive-growing farms in Greece, 1999–2003
| Absolute value | Percentage | |
|---|---|---|
| TFP growth | 1.5777 | 100.0 |
| Technical change | 1.2109 | 76.7 |
| Neutral | 0.9413 | 59.7 |
| Biased | 0.2696 | 17.1 |
| Scale effect | −0.1988 | −12.6 |
| Conventional inputs | −0.2060 | −13.1 |
| Pesticides | 0.0072 | 0.5 |
| Technical efficiency change effect | 0.4865 | 30.8 |
| Autonomous changes | 0.2330 | 14.8 |
| Aridity index | 0.1308 | 8.3 |
| Extension services | 0.1227 | 7.8 |
| Damage-control effect | 0.0791 | 5.0 |
| Abatement effectiveness effect | 0.0839 | 5.3 |
| Initial pest infestation effect | −0.0045 | −0.3 |
| Spillover effect | −0.0003 | −0.0 |
| Absolute value | Percentage | |
|---|---|---|
| TFP growth | 1.5777 | 100.0 |
| Technical change | 1.2109 | 76.7 |
| Neutral | 0.9413 | 59.7 |
| Biased | 0.2696 | 17.1 |
| Scale effect | −0.1988 | −12.6 |
| Conventional inputs | −0.2060 | −13.1 |
| Pesticides | 0.0072 | 0.5 |
| Technical efficiency change effect | 0.4865 | 30.8 |
| Autonomous changes | 0.2330 | 14.8 |
| Aridity index | 0.1308 | 8.3 |
| Extension services | 0.1227 | 7.8 |
| Damage-control effect | 0.0791 | 5.0 |
| Abatement effectiveness effect | 0.0839 | 5.3 |
| Initial pest infestation effect | −0.0045 | −0.3 |
| Spillover effect | −0.0003 | −0.0 |
Note: The time invariant farm-specific characteristics are not taken into account in the technical efficiency change effect.
Among the conventional sources of TFP growth, technical change is found to be the most important as, on average, it accounted for 76.7 per cent of annual TFP growth. As it can be seen from Table 6, the neutral component of technical change is its driven force. On the other hand, the scale effect is negative due to the presence of decreasing returns to scale and a growing aggregate input. The growth of the aggregate input is mainly due to the growth of the conventional inputs as pesticides used had decreased in the period under consideration. However, neither the weight (i.e. output elasticity) nor the decrease in the use of pesticides was enough to outweigh the growth in conventional inputs and thus the scale effect had a negative impact on TFP growth. Technical efficiency changes were the second most important source of TFP growth after technical change and it accounted for 30.8 per cent of annual TFP growth. The effect of positive technical efficiency changes indicates movements towards the frontier over time. As it can be seen from Table 6, all time-varying farm-specific characteristics as well as the passage of time have contributed positively to technical efficiency changes and thus to TFP growth.
On the other hand, abatement effectiveness is by far the most important source of growth among the components of the damage-control effect. The estimated average annual rate of abatement effectiveness was 1.48 per cent and it has contributed 0.084 points of the 1.58 per cent annual growth of TFP, which accounts for 5.3 per cent of its annual growth rate. In contrast, both the pest population and the spillover effect had a negative impact to TFP growth due perhaps to favourable conditions for pest reproduction and the throughout decrease in the use of pesticides by almost all farmers in the sample. Even though the existence of these two components of the damage-control effect cannot be challenged on statistical grounds (Table 4), their combined impact on TFP growth is rather small as together they accounted for only −0.3 per cent of annual TFP growth.
Finally, the overall contribution of pesticides to TFP growth is estimated at 5.5 per cent. This represents the sum of the damage-control effect and the proportion of the scale effect associated with the use of pesticides. The latter has a positive impact on TFP growth as the use of pesticides was declined under decreasing returns to scale. This implies that increases in the use of pesticides, even when are effective in killing pests, would not result in TFP gains if farm size is greater than that maximising ray average productivity. In addition, the proportion of the scale effect associated with the use of pesticides more than offset the negative impact of the pest population and the spillover effects.
6. Concluding remarks
This paper develops a theoretical framework for decomposing TFP growth by taking explicitly into consideration the indirect impact that pesticides have on farm output. Recognising the damage-control nature of pesticides may correct some biases in the measurement and decomposition of TFP related to the overestimated output elasticity of pesticides when it is modelled as an output-expanding input. In the proposed framework, TFP changes are decomposed into the effects of technical change, scale economies, changes in technical efficiency and the damage-control effect. The latter consists of three distinct elements: that due to changes in the level of initial pest infestation, the spillover effect arising from neighbours' use of pesticides and the effect associated with changes in abatement effectiveness.
The model is applied to a panel of olive-growing farms in Crete, Greece, during the 1999–2003 period. The empirical results suggest that technical change was the main source of TFP growth, followed by the effect of technical efficiency changes. The damage-control effect, on the other hand, accounted for a small portion (5 per cent) of TFP growth. The small contribution of the damage-control effect may be specific to the peculiarities of olive tree cultivation and definitely does not imply that it can be neglected without making any difference. To properly decompose the sources of TFP changes, we should explicitly consider the preventive nature of pesticides and thus account for the damage-control effect, regardless of its magnitude in each study case.
Acknowledgements
We would like to thank the seminar participants at the Department of Economics in the Swedish Agricultural University (SLU) and in the University of Cyprus, two anonymous reviewers and one of the editors, Paolo Sckokai, for helpful comments and suggestions. This research has received funding from the European Community's Seventh Framework Programme (FP 7/2008-2011) under grant agreement number 212120.
References
Nevertheless, Hennessy (1997) noticed that even in this case the omission of the initial pest population results in an identification problem, as low realised farm output may be due either to a high level of initial pest density or to a low level of pesticide use.
Underwood and Caputo (1996), using a dynamic farm model, found that the effects of farm programmes (i.e. land set aside, decoupled payments) and pesticide taxes depend crucially on the returns to scale of the output abatement function.
Even with the general assumption that the marginal damage effect of damage agent is non-negative (∂g/∂b ≥ 0), the sign of ∂g2/∂2b is underdetermined. Nevertheless, the damage function is often assumed concave.
In the empirical model, we have also tried the interactive specification of Saha et al. (1997) but it did not work satisfactorily.
It is also assumed that the use of pesticides by farmers in the area does not induce phytotoxicity on-farm.
Chambers and Lichtenberg (1994) demonstrated that separability between damage-control inputs and conventional inputs implies conditional additivity of profit function. Saha et al. (1997) statistically examined the existence of separability between partitions of inputs and they found that the hypothesis of weak separability between land, machinery and miscellaneous inputs and damage-control input use was maintained. They, however, reject that hypothesis between fertilisers and pesticides. Since our data do not provide detailed information on fertiliser applications, we maintained the weak separability assumption which is common to damage-control econometrics.
Notice, however, that time-invariant farm-specific characteristics have no impact on TFP growth.
It is also zero for stationary pest populations.
We have tried to estimate a more flexible translog production frontier but the econometric estimates provided a poor fit of the underlying production technology with the current data set.
Even though this formulation has been used in numerous applications in agriculture and elsewhere, it has recently been criticised for, among other things, not being able to separate heterogeneity and inefficiency (e.g. Abdulai and Tietje, 2007).
We have not included capital as input in for the following reason: the production of the olive-growing farms in the sample is rarely mechanised and their capital consists mainly of their trees. Having, however, no data on tree age and other qualitative aspects, we were only able to construct a broad estimate of capital stock based on the number of trees in each farm, which, for agronomic reasons, is limited to more or less 36 per 0.1 ha. This in turn makes capital and land highly collinear in the econometric estimation and thus we do not lose much information if we keep only one of them, which in this case was land.
The estimated parameters of the augmented model are not presented here but are available upon request.
Generalised likelihood ratio test statistic is computed as λ = −2{ln L(H0) − ln L(H1)}, where L(H0) and L(H1) denote the values of the likelihood function under the null (H0) and the alternative (H1) hypothesis, respectively.
The test statistic computed as (with m3 and m2 being the third and second moments of the residuals and b1 the coefficient of skewness) is 2.124, well above the corresponding critical value at the 5 per cent level of significance (0.298).
