2.1. Data
Data were collected from 1 June 2017 to 7 August 2017 with a sample covering 136 greenhouse tomato and pepper farms from 22 villages, 11 municipalities, and 7 regions of Kosovo. Production information was obtained through face-to-face interviews with farmers. To facilitate the data collection process, two research surveys were developed to gather information from the field. The first survey was targeted toward the greenhouse tomato production in the seven regions, which includes Prishtina, Ferizaj, Gjilan, Prizren, Gjakova, Peja, and Mitrovica. The second survey was developed for the greenhouse pepper farms and was executed in all of the same regions excluding Prizren because of limited greenhouse pepper production in this area. Each survey included 47 questions and was separated into four survey modules. Both of the surveys contained three modules that elicited information about the respondent, characteristics of the greenhouse, and assessed producer needs. A fourth survey module was designed specifically for tomato or pepper production, respectively. All farmers interviewed were growing at least tomatoes or peppers as their cash crop. Some of those same farmers also produced a few rows of these crops or other vegetables for home consumption.
2.2. Data Envelopment Analysis (DEA)
The purpose of this analysis was to determine technical efficiency (TE), pure technical efficiency (PTE), and scale efficiency (SE) of the greenhouse tomato and pepper farms. An additional goal of this study was to understand factors that influence TE scores. Therefore, an input-oriented DEA coupled with linear regression were used to compare input efficiency use between farms and to explore what farm-level variables could predict TE. As a mathematical technique, DEA solves performance evaluation problems [
7] and quantifies input efficiency use. The input-oriented Banker-Charnes-Cooper (BCC) [
8,
9] and Charnes-Cooper-Rhodes (CCR) [
10] models were used. After characterizing TE and SE from these models, TE scores received from the greenhouse tomato and pepper CCR models were used as outcomes in additional analyses to understand farm-level characteristics that impact efficient input use. All analyses were performed in R, which is a programming language and software for statistical analysis.
Using the BBC model, we let j = 1, 2, ..., n, index the greenhouse tomato and pepper operations. Each operation realizes yields, y
rj (r = 1, 2, ..., s), using observed levels of m farm inputs, x
ij (i = 1, 2, ..., m). In addition, ε denotes a small value that penalizes slack terms
and
to facilitate model solutions and prevent a classification of an inefficient decision making unit (DMU) as efficient. A small and positive value for ε may work in many instances [
11]. The BCC model is presented as follows.
subject to
Yield is the output for both greenhouse tomato and pepper farms because farmers’ primary production decisions can be based on potential future yields [
12]. Efficient operations of farms can also lower per unit production costs and lead to yield increases [
13]. The inputs used in the models include insecticide, labor, greenhouse area in square meters (m
2), greenhouse value in euros, and the use of artificial and organic fertilizers at different stages of greenhouse production.
The rationale for measurement and inclusion of these variables in the model is as follows. In Kosovo, often greenhouses covering larger areas correspond to greenhouses that have improved designs and structures. These greenhouses may have, on average, higher volumes of production and may be more efficient. Therefore, the variable greenhouse area in m
2 was included. Among pesticide use in Kosovar greenhouse tomato and pepper production, insecticides may be used more regularly than herbicides and fungicides. Similarly, chemical pest control measures can be costly in Kosovo but may be highly beneficial for production [
14]. Labor was included as an input because two studies [
4,
15] presented evidence that labor is often overused in tomato production. Since this research suggests that labor is prone to overuse in greenhouse tomato production, it was hypothesized that labor could also be overused in pepper production.
The literature noted that fertilizer was an important input to include in the evaluation of TE [
15,
16]. One consideration is whether artificial or organic fertilizers may have a greater impact on tomato and pepper yields. Although several studies suggested that organic fertilizers may do more to increase productivity [
17,
18,
19], there is evidence that the tomato yield was higher with the use of mineral fertilizers than with organic fertilizers [
20]. To explore how fertilizers impact greenhouse tomato and pepper input efficiency use, this study included the planting phase of organic and artificial fertilizers and the flowering phase among crystalline and artificial fertilizers. In discussions with Kosovar experts, the value of greenhouses in Kosovo varied greatly between government program grantees and non-grantees. Grantees had greenhouses of higher euro value. Therefore, this study considered the greenhouse value in euros as an input for evaluating input efficiency use among greenhouse tomato and pepper farms.
The BCC model differs from the CCR model in that it includes an additional constraint on the intensity variables
[
9]. This can permit the measurement of efficiency relative to a variable return to scale (VRS) technology. A PTE score is represented by
. The CCR model measures efficiency relative to a constant return to scale (CRS) technology and
corresponds to a TE score. This model uses the same inputs and outputs explained previously in this study for evaluating greenhouse tomato and pepper farms’ TE. In a linear programming framework, DEA as a non-parametric method was used to calculate SE. The evaluation of the CCR and BCC models helped us define SE as
. With the use of these notions, this study may demonstrate the decomposition of efficiency, which is shown below.
The purpose of the decomposition was to show whether the source of inefficiency was caused by PTE, by SE, or by both [
21].
2.3. Regression Analysis
This study contained two datasets that were used separately for greenhouse tomato and pepper farms. It was possible to analyze the efficiency scores derived from DEA through a linear regression model. An equal number of variables was considered for estimating the linear regression model for both the greenhouse tomato farms and the greenhouse pepper farms (
Table 1).
Using the variables above from
Table 1, we examined the average effect of farm characteristics on the TE scores of tomato and pepper greenhouses. The TE scores derived from the CCR model were used as the outcome measure in this regression.
Following the model, this study called the technical efficiency as the dependent variable for a given tomato or pepper greenhouse. The random error term for a greenhouse was represented by .The first variable used was . It was useful to explore which group of farmers were more efficient in the use of inputs including those who expressed their need to participate in crop nutrition training and those who did not.
Energy efficiency improvements are vital [
22]. The government of Kosovo has prioritized the energy sector by emphasizing the improvement of electricity generation capacities. Over the years, a steady increase in electricity production has been noted in Kosovo. However, challenges remain in the effort to gain an efficient use of electricity [
23]. Enterprises and farms in Kosovo may examine a cost-effective way to consume energy. There may be variations between greenhouse farmers who used electricity and greenhouse farmers who used fuel as their power source at a farm level. This power source may have an impact on a farm’s TE. Electricity remains a costly source of power in Kosovo [
24]. Farmers focus on minimizing costs where the source of power often accounts for a large share in the vegetable production costs. Considering this, we used the indicator variable
. Fruit yields can increase when the greenhouse plants are arranged correctly and when there is a minimization of gaps between plants and rows [
25]. To achieve more greenhouse tomato or pepper yields, farmers may increase the number of rows more than may be efficient. Likewise, a discrepancy in the number of rows per greenhouse can impact how each farmer uses inputs in the production process. It is of interest to find an appropriate number of rows in the greenhouse, which may affect how inputs are allocated. Therefore,
was included.
An issue reported from the greenhouse farmers is the low price received per kilogram (kg) of the produce. In this study,
is one variable that may explain this issue in part. Farmers often can have high price expectations if they noticed that there were high wholesale prices in the market from the previous harvesting season [
26]. When the price is low from the vegetable wholesalers, greenhouse farmers may have to contract their gross profit margins. Actually, farmers can be forced to market their produce at lower prices to avoid the risk of not selling. In addition, farmers’ ability to sell their produce may be influenced from the wholesale price of tomatoes that tends to fluctuate. [
27]. To test if the variable including the farmer’s wholesale selling price has an impact in the optimal use of inputs, this study included it in the models. Similarly, prices and varying yields can influence farm incomes [
28]. Even off-farm income was found to have a positive effect on revenue risk [
29]. The use of
tested whether farmers with and without an external source of revenue differed in the optimal use of inputs. Off-farm income may even substitute income losses that occurred in the farm [
29,
30]. However, there may not be sufficient evidence to conclude why some farmers rely on off-farm income and others do not [
30]. It can be expected in this study that farmers who have an external source of revenue could rely less on on-farm revenue.
Vegetable farms growing two or more crops were found to have less usage of water, diesel, and electricity [
31]. In this aspect,
was another variable used in the models to understand how growing other crops over the course of a season impacts the efficient use of inputs. In Kosovo, there may be a mixture of farmers growing greenhouse tomatoes or peppers as a single crop and those who may have other crops with tomatoes or peppers in the same greenhouse. Large-sized and mixed farms tend to have high efficiency [
32]. The variable
can be important given that product prices in agriculture have a high tendency to vary [
12]. Farmers may choose to sell directly to the retailers or consumers through farmers’ markets [
33]. High quality peppers can achieve premium prices in the market [
34] while conventional tomatoes relative to organic tomatoes may not reach premium prices in the market [
35]. There is a belief in Kosovo that farmers selling greenhouse peppers at the farmer’s market may receive higher prices per kg compared to the greenhouse tomato farmers. Whether the farmer market price influences the efficient use of inputs was important to explore.
The variable
was included because the amount of water applied on crops has a clear tendency to affect yields [
36] and an efficient use of irrigation would rely on the design of the irrigation system and its management [
37]. For example, an implication of a limited irrigation time could suggest that farms in Kosovo’s regions with lower than average well depths may be less likely to irrigate during the flowering season when faced with increased levels of water scarcity. When the well depth is large and there is an increase in irrigation effectiveness, a potential to grow yields is possible [
38]. This study expects that it could be possible to test if well depth is likely to influence the optimal use of inputs. The model has
as a variable that may provide an understanding of the quality of a farmer’s irrigation system. Inadequate irrigation of the vegetable crops because of the old irrigation equipment can constrain input efficiency use. Despite the wide presence of the drip irrigation systems in Kosovo, the frequency and amount of irrigation needs improvement [
39]. The greenhouse tomato and pepper farmers may have irrigation systems with varying euro values. Therefore, it was essential to use this variable for the greenhouse tomato and pepper input efficiency use related to regression analyses.
Studies in agriculture have found education to positively impact higher levels of TE [
40,
41]. However, education may not be significantly correlated with efficiency [
42]. Therefore, the variable
could test whether education has an impact on the greenhouse tomato and pepper TE. The use of
is relevant since most of the farm work is performed by the farmer’s family members [
14]. The number of family members may dictate the intensity of family labor use. The limited literature in Kosovo on the impact of family members in the production of greenhouse tomatoes and peppers could allow this study to examine any potential influence from the variable.