Evaluating the Technical Efficiency of Dairy Farms Under Technological Heterogeneity: Evidence from Lithuania

Abstract

The European Union’s (EU) Common Agricultural Policy aims to promote sustainable farming practices that ensure the responsible use of natural resources, safeguard biodiversity, and uphold higher animal welfare standards. One pathway to achieving these objectives is through the encouragement of extensive farming. However, the dairy sector in EU countries as well as in Lithuania has shown a clear trend toward intensification. The aim of this study was to assess the technical efficiency (TE) of dairy farms employing extensive and intensive technologies. TE was evaluated using Data Envelopment Analysis (DEA) combined with meta-frontier analysis, which accounts for technological heterogeneity. Prior to the efficiency estimation, farms were grouped into two distinct categories—intensive and extensive—using the k-means clustering algorithm. The empirical results show that extensive dairy farms in Lithuania are smaller in land area and livestock units, rely more on internal resources, and exhibit lower productivity compared to intensive farms. Intensive farms achieved higher technical efficiency, narrower technological gaps, and more optimal scale efficiency, indicating superior resource management. The weaker performance of extensive farms is attributed to both less advanced technologies and production inefficiencies.

1. Introduction

The EU agricultural policy has set a goal to promote farming practices focused on the sustainable use of natural resources, biodiversity conservation, and the implementation of higher animal welfare standards. This policy orientation is outlined in the Farm to Fork Strategy [1] and the Biodiversity Strategy [2]. One way to achieve these goals and reduce environmental pressure is by promoting extensive farming. EU policy supports a range of measures that encourage extensive farming, such as eco-schemes, support for organic farming, extensive grazing, and non-productive investments. Extensive dairy farming refers to a milk production system characterized by low livestock density and a predominant reliance on home-grown feed [3].

In contrast, statistics show that European dairy farms have been moving toward more intensive production systems [4,5]. The dairy sector in EU countries has experienced a trend toward intensification, which became particularly pronounced after the abolition of milk quotas in 2015 [6,7]. An analysis of data from the Dairy Farm Structure Survey 2010–2020 (Eurostat, the most recent relevant survey) shows that the number of farms with dairy cows decreased by 51.9% between 2010 and 2020, while the average number of cows per farm in the EU countries doubled, from 12.6 cows in 2010 to 25.3 cows in 2020 [8,9].

These changes were driven by a stronger market orientation of agricultural policy, a change in the direct support model, and the abolition of export subsidies, which opened up the sector to global markets [10]. Global trade operates at world market prices, while production costs remain local, meaning that costs determine which producers are likely to survive in the long term [11]. As a result, in order to remain competitive, farms are forced to increase their production volumes, leading to an increase in average farm size. Many smaller farms, unable to remain profitable, have been forced to exit agriculture and transition into other sectors of the economy [12,13,14,15]. Modern dairy farms now typically keep larger herds, achieve higher milk yields, and use greater quantities of compound feed. This trend is particularly evident in larger dairy farms, where economies of scale enable profitability despite continuously rising feed costs [16]. However, the concentration of cows and intensive milk production also pose environmental challenges, including biodiversity loss and water pollution.

Farming, like any other economic activity, is driven by the objective of generating profit. Therefore, in order to evaluate whether promoting extensive farming can be economically viable, it is necessary to assess and compare the economic performance of dairy farms employing different technologies. Technical efficiency (TE) is a core indicator of a farm’s economic performance and a valuable tool for guiding improvements in production processes [17]. For this reason, our study focuses on evaluating the TE of dairy farms using different technologies. Understanding these differences is essential for policymakers, farmers, and researchers seeking to develop appropriate strategies for sustainable dairy production.

According to the classical definition by Farrell (1957) [18], a farm is considered technically efficient if it produces the maximum possible output given the amount of inputs and the technology used. Studies on TE are important because inefficiency increases production costs and reduces the competitiveness of farms.

Dairy farming systems can broadly be categorized into intensive and extensive technologies, each with distinct characteristics that affect productivity, efficiency, and environmental impact. The existing scientific literature addresses the issue of production heterogeneity by ex ante grouping the sample based on a priori knowledge or well-founded assumptions regarding technological differences. In the second stage, frontier functions are then estimated for each homogeneous group of farms, which are assumed to share the same production technology. Alternatively, models that allow for the identification of latent (unobserved) heterogeneity are applied within the framework of stochastic frontier analysis (SFA) [19,20,21].

In our study, we applied Data Envelopment Analysis (DEA) combined with metafrontier analysis to assess the TE of dairy farms. DEA is a non-parametric method that uses linear programming to construct a data envelopment surface and measure each observation’s distance (inefficiency) from the frontier. Prior to conducting the efficiency analysis of dairy farms using the DEA meta-frontier approach, the farms were classified into two clusters—intensive and extensive farms using the k-means clustering algorithm. Battese et al. (2004) [22] and O’Donnell et al. (2008) [17] introduced the modern meta-frontier production–function model. Meta-frontier analysis can assess the technology gap, which quantifies the extent to which a specific group’s frontier deviates from the metafrontier, thereby highlighting potential directions for technological improvement [17]. The meta-frontier model takes into account technological differences between groups, but does not cover specific farm constraints within groups. To overcome this limitation, future studies would benefit from applying conditional DEA or including environmental variables to more accurately reflect farm efficiency under such constraints.

The aim of this study is to apply efficiency assessment methods to dairy farms operating under different technological systems. The novelty of the research lies in its application to two distinct technological groups—intensive and extensive—within the Lithuanian dairy farms, which has not been previously examined from this perspective. This distinction is particularly relevant in the context of the EU Common Agricultural Policy (CAP) 2023–2027, which emphasizes the need to increase the sustainability, competitiveness and resilience of agriculture [1]. Differentiating the performance of farming systems contributes to the CAP’s objective of tailoring policy instruments such as eco-schemes, targeted support for less-favored areas and rural development interventions.

The study not only assesses technical efficiency within each group, but also examines how performance patterns differ across technological systems. These insights are essential to improve the accuracy and effectiveness of policy instruments, in particular when reconciling productivity with environmental and social objectives.

The empirical analysis focuses on Lithuanian dairy farms. The TE of Lithuanian dairy farms has been assessed under the assumption that farms operate under a common production technology [23,24]. The results indicated that Lithuanian dairy farms exhibit a considerable degree of performance heterogeneity. These results underscore the structural diversity within the extensive farming group, challenging the assumption that such farms form a homogeneous category. The findings emphasize the importance of considering structural constraints when evaluating farm performance. Moreover, they highlight that a one-size-fits-all policy approach may be inadequate, given the diversity of production conditions and capabilities. Future research would therefore benefit from applying conditional DEA or meta-frontier models that take into account environmental and structural constraints in performance evaluation.

The article is organized as follows. Section 2 provides a literature review aimed at identifying the indicators used to distinguish heterogeneous technologies in dairy farms. Section 3 describes the methods used to assess TE, considering technological heterogeneity. The results are discussed in Section 4. Finally, the last section draws conclusions on our empirical results.

2. Literature Review

According to the classical definition by Farrell (1957) [18], a farm is considered technically efficient if it produces the maximum possible output given its input levels and the technology used. Studies on TE are important because inefficiency increases production costs and reduces the competitiveness of farms [25].

The traditional method used to evaluate efficiency indicators is based on the assumption that a common production technology applies to all farms in the sample. However, farms operate under different conditions; farms located in different regions face varying production environments, and technologies may differ due to variations in resource availability and production methods. Therefore, incorporating technological heterogeneity into the analysis should reduce distortions in the assessment of farm performance [26]. Farms choose from different combinations of inputs and outputs based on their specific production conditions and circumstances [17].

Therefore, to obtain more objective research results, farms are grouped according to certain characteristics, based on prior knowledge or well-grounded assumptions about technological differences, so that the selected group of farms is more homogeneous. Then, in the next stage, the analysis is conducted within these homogeneous groups, as farms within the same group are comparable and efficiency comparisons within the group are valid [27].

The selection of criteria to obtain more homogeneous farm groups depends on the research objectives and the methods used. Many studies have analyzed the relationship between TE across different groups of farms based on regional classification, assessing the impact of varying environmental conditions on farm TE. This approach is frequently applied to evaluate farms both at the national level [19,28,29] and to compare TE differences between countries [30,31].

The production technologies used in dairy farming represent another important factor when assessing TE. In research, a clearly defined characteristic may be used to identify the type of production technology—for example, studies comparing the TE of organic and conventional dairy farms [32]. Alternatively, technologies may be differentiated based on selected criteria. In general, production intensification is reflected in a higher stocking rate (more dairy cows per hectare), the use of genetically improved dairy cattle (higher milk yield per cow, milk yield per hectare), and a greater proportion of concentrates in the feed ration [33,34,35]. In order to avoid potential endogeneity problems, studies seeking to assess the nature of intensification typically use indicators describing cattle density or feed use [3]. Ahikiriza et al. (2021) [27], Ojo et al. (2020) [36] used the external input cost criterion to differentiate between farm technologies. Based on the assumption that farms with different external input strategies employ different production technologies, technologies were classified according to input intensity—low-input and high-input technologies—and the efficiency of these farm groups was evaluated accordingly. Latruffe et al. (2023) [20] used livestock density, the proportion of feed area in total agricultural land, and the proportion of leased land in total agricultural land to distinguish between intensive and extensive farms.

In order to assess the economic performance of farms or TE more accurately, studies also use a two-level classification of farms, where farms are first grouped based on a predefined characteristic, and then additional criteria are used within each group to identify the type of technology applied. For example, Dakpo et al. (2022) [21] divided dairy farms into two groups depending on whether they participated in agri-environmental schemes, and within each group, they assessed the TE of intensive and extensive farms. In another study, Dakpo et al. (2021) [37] analyzed three types of grazing livestock farms in France and identified two farm classes: intensive and extensive. Similarly, Garcia-Covarrubias et al. (2023) [38] examined the TE of Irish dairy farms using automated technologies, classifying them into two categories: smaller, less intensive farms and larger, more intensive farms.

Various clustering methods can also be applied to identify homogeneous farm groups. For instance, Sauer (2011) [39], in assessing the performance of Danish dairy farms using a latent class model, grouped farms based on four types of features to distinguish between technologies. Gonzalez-Mejia et al. (2018) [40], evaluating the impact of intensification in dairy farms, used indicators from seven groups to cluster the farms. Similarly, Alvarez et al. (2008) [33] used the k-means method to cluster farms into extensive production technology and intensive production technology in order to assess cost differences in milk production.

In studies conducted at the EU level, clustering is commonly used to address farm heterogeneity by grouping dairy farms across EU countries [13,41,42]. For example, Poczta et al. (2020) [13] applied the Ward hierarchical method to classify EU dairy farms into five clusters, based on selected criteria: share of hired labor in total labor inputs; share of leased agricultural land in total agricultural land; total assets minus the value of land, permanent crops and production quotas; agricultural land area per full-time employee; number of dairy cows; share of livestock in fixed asset value; and livestock density, to evaluate their economic performance. Náglová & Rudinskaya (2021) [42] divided dairy farms into four clusters based on their physical size (number of livestock units per farm) and economic size (standard output per farm), and then assessed the TE of each cluster.

The validity of applying clustering depends on solid evidence confirming the heterogeneity of farms. When clustering is applied, farms are grouped based on shared characteristics to ensure homogeneity within groups and heterogeneity between groups. Within-group analysis enables the comparison of farms with similar counterparts, ensuring internal consistency and result comparability. In contrast, comparisons across clusters are expected to show significant differences, as farms are deliberately separated based on differing operational environments and technologies. This highlights the importance of selecting appropriate criteria when forming clusters, as inappropriate choices may lead to misleading results [17].

Table 1. Indicators used for grouping individual dairy farms to identify technological heterogeneity.

Indicator	Method	Author
Cows per hectare of land, purchased feed (purchased concentrate) per cow	Latent class model	(Alvarez & Arias, 2015 [3]; Alvarez & Del Corral, 2010 [35])
Milk per cow, milk per hectare, feed per cow, and cows per hectare	Cluster analysis	(Alvarez et al., 2008 [33])
Livestock density (stocking rate, no. total livestock unit (LU) per hectare of forage area); the ratio of fodder area to UAA; the share of the rented area to UAA	Latent class model	(Latruffe et al., 2023 [20])
Stocking rate (no. dairy cows per ha); feeding intensity (purchased feed per cow)	Latent class model	(Stetter et al., 2023 [43])
Stocking rate (no. LU per ha of UAA); share of permanent grassland in the UAA	Latent class model	(Dakpo et al., 2022 [21])
Intensive/extensive nature of production: stocking rate (no. dairy cows per ha), fodder quantity per cow; Organic/conventional systems: share of organic milk revenue in total revenue or chemicals per hectare; The input intensity of production: labor per cow, capital per cow; Specialization dairy: Share of milk revenue in total revenue.	Latent class model	(Sauer, 2011 [39])
Milk production specifically (no. of dairy cows, milk yield, concentrate feed per LU, fodder per LU, milk premium); Intensity and specialization of livestock production (dairy stocking density, livestock density, dairy fraction (share of dairy cows in LU), labor intensity (annual worked hours per farm area), and fodder per LU); Grazing prevalence (fodder area/grass area, maize area/grass area); Farm structure for animals (non-cash crop area/UAA, grass area/UAA); Production area (UAA/Farm Area); Tenure (owner occupied area/UAA); Replacement rate (heifers/dairy cows).	Cluster analysis	(Gonzalez-Mejia et al., 2018 [40])
Livestock density (LU per hectare); location in the mountainous region; indicator whether a tie-up barn or free-stall housing system is used; indicator about silage-free production.	Latent class model	(Renner et al., 2021 [44])
Labor intensity (labor hours per LU); farm stocking rate (LU per UAA); specialization dairy (share of dairy gross output in the farm gross output).	Latent class model	(Garcia-Covarrubias et al., 2023 [38])
Input intensity—external input costs per grazing livestock unit or per dairy cows. Farms with high input costs and those with low input costs, corresponding to high and low input technologies, respectively.	Mathematical partitioning	(Ahikiriza et al., 2021 [27]; Ojo et al., 2020 [36])
Farm intensity: stocking rate (no. of livestock units per ha of UAA); share of permanent grassland in UAA; capital intensity (the ratio of fixed assets per labor unit); Environmental practices (the amount of CAP agri-environmental subsidies per hectare of UAA); Weather conditions (average daily effective rainfall (in mm) and temperature (in degrees Celsius). External factors (farm location in LFA).	Latent class model	(Dakpo et al., 2021 [37])

presents the indicators used by researchers in their studies to divide the farm sample into distinct groups, with the aim of forming more homogeneous clusters and evaluating the economic performance of each group. The literature review shows that the selected indicators vary and depend on the specific research objectives and methodological approaches used.

All studies identified significant differences between the identified groups of farms, leading to the conclusion that distinguishing homogeneous technologies and farm types is an important factor in evaluating economic performance and in developing more targeted operational recommendations. While it is challenging to account for all existing differences, considering at least some of them is essential in order to obtain reliable results [17,22].

Strategic documents adopted by the European Commission between 2019 and 2024—including the European Green Deal [45], the Farm to Fork Strategy [1], the EU Biodiversity Strategy [2], the Circular Economy Action Plan [46], and the EU Nature Restoration Law [47]—aim to restructure agriculture in a way that promotes more extensive farming practices, conserves and protects natural resources, halts soil degradation, and restores biodiversity. In response to these EU policy goals, this study aims to assess the TE of dairy farms by classifying them according to the intensity of their production technologies and taking into account environmental pressures such as land use intensity, nutrient loads and grassland degradation. Farms were clustered using indicators identified in the literature as related to the extensive or intensive nature of production.

3. Materials and Methods

In the scientific literature on agricultural research, two main methods are widely used to assess TE in farming [48]: Stochastic Frontier Analysis (SFA) [3,19,20,21,32,35,37,38,44] and Data Envelopment Analysis (DEA) [23,27,31,43,49,50,51,52,53].

SFA is a parametric method that can accommodate noise, such as measurement errors caused by disease, weather conditions, and the inefficiency component. Its application is appropriate in cases where a well-established functional relationship exists, and the production function is known in advance [22,35]. DEA is a widely used non-parametric method for evaluating the relative efficiency of farms. It ranks a set of decision-making units (DMUs) based on their performance by solving a linear programming problem to calculate TE. DEA identifies which units operate efficiently and which exhibit potential for improvement by benchmarking them against the best-performing peers. The efficiency frontier is constructed as a piecewise linear envelopment surface [17,27,52].

In this study, DEA was selected due to its flexibility, its non-reliance on a predefined production function, and its suitability for the relatively small sample size of farms. Moreover, DEA can be effectively integrated with meta-frontier analysis, allowing for the assessment of TE across heterogeneous technological environments.

To assess farm TE while accounting for farm heterogeneity, various analytical methods have been developed, among which the most popular are meta-frontier analysis [17,19,22,27,29] and latent class models [20,22,27,35,37,38].

Latent class models partition the sample and estimate TE of each group in a single stage. Each farm can be assigned to a specific group based on the estimated probabilities of class membership, determined using several selected criteria [35]. However, the latent class model cannot be applied in combination with DEA; therefore, in our study, we used the meta-frontier approach.

DEA combined with meta-frontier analysis is a two-stage method. In meta-frontier analysis, farms are first grouped according to different technologies, and the relative TE of farms operating under distinct technologies is then calculated. Clustering enables a more targeted analysis within each group; however, it does not allow for direct efficiency comparisons across groups. This analytical gap is addressed by the meta-frontier approach. The meta-frontier concept is based on the assumption that, although DMUs may operate under different conditions and employ different technologies, there exists a hypothetical best-practice frontier that could be achieved under optimal conditions [52]. Meta-frontier analysis can assess the technology gap, which quantifies the extent to which a specific group’s frontier deviates from the meta-frontier, thereby highlighting potential directions for technological improvement [17].

Prior to conducting the efficiency analysis of dairy farms using the DEA meta-frontier approach, the farms were classified into two clusters using the k-means clustering algorithm. In this method, the number of clusters must be defined by the researcher. The objective of the clustering was to group farms according to the extensive or intensive nature of their production practices, considering the environmental pressures of their production practices. Based on the literature review, two indicators were chosen: stocking density and self-sufficiency in feed—the share of self-produced feed as a proportion of the total feed used on the farm. A correlation analysis was carried out to select the indicators in order to avoid strong multicollinearity and ensure the reliability of the analysis. Livestock density and feed self-sufficiency are indicators for assessing the environmental sustainability of farms (related with subthemes: land use pressure, nutrient load and pollution risks, grassland degradation and circularity, dependence on other productive sectors [54,55]). A decreasing land area per cow indicates a trend towards production intensification and reflects increased exploitation of natural resources [56]. Moreover, land availability per cow serves as an indicator of higher animal welfare standards on farms, such as access to outdoor areas or grazing opportunities. The indicator of self-sufficiency in feed can also be considered as an indicator of circularity, as it reflects one of the principles of circularity—minimizing external resource input [57]. The cluster analysis identified two distinct clusters, interpreted as representing intensive and extensive farming systems. To assign each farm to a cluster, the k-means algorithm calculated the similarity between observations using Euclidean distance, with all variables standardized to ensure comparability.

In the next step, the DEA method was applied to evaluate the efficiency levels of farms within each cluster, followed by a meta-frontier analysis. The DEA meta-frontier analysis was conducted using R 4.4.2 software, estimating two group-specific frontiers and a single meta-frontier. These were estimated using one output—total farm output in EUR—and five inputs:

• Utilized agricultural area (UAA, X₁) comprises both owned and rented area measured in hectares.
• Labor (X₂), measured in hours worked by annual work units (AWU), includes both family labor and hired labor.
• Herd size (X₃) measured in livestock units (LU).
• Intermediate consumption (X₄) includes specific costs for agricultural production (seeds, fertilizers, crop protection products, crop insurance for crops, purchased concentrates, purchased coarse fodder, farm use of non-fodder crops, specific forage costs, milk herd renewal costs, the milk levy and other specific livestock costs (veterinary, etc.)) + non-specific costs: upkeep of machinery and buildings, power (fuel and electricity), contract work, taxes and other dues (excluding the milk levy), taxes on land and buildings, insurance for farm buildings and other direct costs.
• Capital assets (X₅)—the value of machinery and buildings at the beginning of the year. The capital assets do not include the value of agricultural land and the value of livestock to avoid double counting.

This input selection was based on the fact that these five components account for the total production costs in dairy farming. This research has used an input-oriented model with the aim of assessing which input variables could be minimized by farmers, in terms of both variable returns to scale (VRS) and constant returns to scale (CRS).

The research methodology is based on the modern meta-frontier production function model developed by Battese et al. (2004) [22] and O’Donnell et al. (2008) [17]. All decision-making units (DMUs, in our case farms) (# = 176) $j$ are divided into two groups (clusters) based on k-means cluster analysis. All farms belong to non-overlapping subsets and $Jg$ is the number of DMUs in each group. The production possibility set for group $g$ will be:

$$T^g=\left\{\left(x, y\right)÷x\ge \sum _{j\in Jg}{\lambda }_{gj }{x}_{g }^j; y \le \sum _{j\in Jg}{\lambda }_{gj }{y}_{g }^j; \sum _{j\in Jg}{\lambda }_{gj }=1; {\lambda }_{gj }\ge 0 \right\}$$

(1)

The set $T^g$ in Equation (1) is the free disposal convex hull of the observed input–output bundles of DMUs from group $g$ and ($x^j, y^j$) are the observed input–output bundle of an individual DMU $j$ in a sample of $Jg$ DMUs in the data. In an input-oriented model, a measure of within-group technical efficiency (${TE}_g$) of DMU $k$ belonging to group $g$ can be given as:

$${TE}_g^k={\theta }_g^k$$

(2)

where ${\theta }_g^k$ solves the linear program in Equation (3):

$$Min {TE}_g^k={\theta }_g^k$$

(3)

This is subject to:

$$\sum _{jϵJg}{\lambda }_{gj }{x}_g^j{\le \theta }_{g }^k{x}_g^k; \sum _{jϵJg}{\lambda }_{gj }{y}_g^j\ge y_g^k; {\lambda }_{gj }\ge 0$$

(4)

The linear programming problem is subsequently solved for each farm within every group. Given the assumption of Variable Returns to Scale (VRS), an additional convexity constraint, as outlined in Equation (5), is incorporated into the model.

$$\sum _{jϵJg}{\lambda }_{gj }=1$$

(5)

The next step is the assessment of the TE of the same DMU $k$ from group $g$ relative to the meta-frontier. The meta-frontier represents the outer envelope encompassing all group-specific frontiers. It is defined by the boundary points of the free disposal convex hull formed from the input–output vectors of all DMUs in the sample. The overall technical efficiency $\left({TE}_G\right)$ of DMU $k$ in group $g$ is calculated as follows:

$${TE}_G^k={\theta }_G^k$$

(6)

where ${\theta }_G^k$ solves the linear program in Equation (7):

$$Min {TE}_G^k={\theta }_G^k$$

(7)

This is subject to:

$$\sum _{g=1}^2\sum _{jϵJg}{\lambda }_{gj }{x}_g^j{\le \theta }_{G }^k{x}_g^k; \sum _{g=1}^2\sum _{jϵJg}{\lambda }_{gj }{y}_g^j\ge y_g^k; {\lambda }_{gj }\ge 0$$

(8)

Given the assumption of VRS, an additional convexity constraint is incorporated into the model (in Equation (9)).

$$\sum _{g=1}^2\sum _{jϵJg}{\lambda }_{gj }=1$$

(9)

In addition to the DEA VRS assumption, TE under the constant returns to scale (CRS) assumption was also calculated within each group. Unlike the DEA VRS model, the CRS model does not include an additional convexity constraint ($\sum _j{\lambda }_{gj}=1$), and therefore, the frontier is a linear envelopment curve, which is appropriate when it is assumed that all farms operate at an optimal scale.

Scale efficiency (SE) was also calculated. SE is defined as the ratio of TE under CRS to TE under VRS:

$${SE}_k={\displaystyle \frac{{TE}_{{CRS}_k}}{{TE}_{{VRS}_k}}}$$

(10)

SE indicates the proportion of efficiency loss that is due to scale inefficiency. An SE value equal to 1 indicates that the farm operates at an optimal scale. SE less than 1 means that there are efficiency losses due to either too small or too large a scale of operation.

The meta-frontier encompasses the production possibility sets of all groups; therefore, ${\theta }_{G }^k\le {\theta }_{g }^k$ for all $k$ and $g$. This means that farmers evaluated against the meta-frontier cannot be more technically efficient than when evaluated against their group-specific frontier.

The Technology Gap Ratio (TGR) quantifies the distance between a group-specific frontier and the common meta-frontier. It is defined as the ratio of a farm’s TE relative to the meta-frontier to its efficiency relative to the group-specific frontier. When these efficiency scores are similar, it indicates that the group’s technology closely approximates the technology represented by the meta-frontier. An overall measure of proximity for group $g$ can be obtained by calculating the geometric mean of the individual TGR values.

For cluster $g$, the overall proximity of the group frontier to the meta-frontier is captured by TGR, defined as:

$$TGR\left(g\right)={\displaystyle \frac{{TE}_G\left(g\right)}{{TE}_g\left(g\right)}}$$

(11)

where ${TE}_G\left(g\right)$ is the average TE of group $g$, measured from the meta-frontier, and ${TE}_g\left(g\right)$ is the average TE of group $g$.

TGR increases as the group frontier approaches the meta-frontier, holding other factors constant, and is bounded above by 1. A TGR value of 1 indicates that the group frontier coincides with the meta-frontier. This suggests that while farms (DMUs) within a group may be highly efficient relative to each other, their efficiency may decline when evaluated against the broader benchmark of all DMUs.

By comparing group-specific frontiers to the meta-frontier, the TGR highlights the level of improvement required for farms using a particular technology to achieve the efficiency levels of those operating on the meta-frontier. A TGR of 1 indicates no technological gap—such farms require no improvement. Conversely, the smaller the TGR, the greater the potential for a farm to enhance its performance to reach the meta-frontier standard.

4. Results and Discussions

The section on results is divided into several parts. The first part describes the dairy sector in Lithuania. Other parts are devoted to the analysis of technical efficiency analysis in relation to selected criteria that may influence efficiency.

4.1. The Dairy Sector in Lithuania

Over the past two decades, Lithuania’s dairy sector has undergone significant transformation. Integration into the European dairy market and exposure to global competition, along with changes in regulatory instruments—such as the abolition of the milk quota system, the discontinuation of export subsidies, the phasing out of direct support, and the introduction of decoupled payments—have driven structural changes in dairy farming.

Table 2. Developments of the Lithuanian dairy sector during the 2004–2023 period.

	Farms		Cows		Milk Production		Milk Yield Per Cow
	Number (Thous.)	Rate (%)	Number (Thous.)	Rate (%)	Thous. t	Rate (%)	kg	Rate (%)
2004	195.2		467.4		1848.7		4176
2005	181.3	−7.2%	462.9	−1.0%	1861.6	0.7%	4312	3.3%
2006	164.6	−9.2%	438.1	−5.4%	1891.3	1.6%	4484	4.0%
2007	142.3	−13.5%	420.6	−4.0%	1936.6	2.4%	4708	5.0%
2008	121.0	−15.0%	395.9	−5.9%	1883.8	−2.7%	4778	1.5%
2009	107.8	−10.9%	380.2	−4.0%	1791.0	−4.9%	4811	0.7%
2010	99.5	−7.7%	357.1	−6.1%	1736.5	−3.0%	4901	1.9%
2011	91.1	−8.4%	345.3	−3.3%	1786.4	2.9%	5026	2.6%
2012	77.6	−14.8%	328.4	−4.9%	1778.1	−0.5%	5227	4.0%
2013	70.6	−9.0%	316.4	−3.6%	1722.3	−3.1%	5315	1.7%
2014	64.4	−8.8%	310.4	−1.9%	1795.1	4.2%	5665	6.6%
2015	60.1	−6.7%	313.5	1.0%	1738.5	−3.2%	5636	−0.5%
2016	53.7	−10.6%	300.6	−4.1%	1627.7	−6.4%	5536	−1.8%
2017	47.1	−12.4%	285.4	−5.1%	1570.7	−3.5%	5601	1.2%
2018	41.4	−12.2%	272.1	−4.7%	1571.8	0.1%	5934	5.9%
2019	36.0	−12.9%	256.7	−5.7%	1551.1	−1.3%	6225	4.9%
2020	30.9	−14.4%	241.8	−5.8%	1491.7	−3.8%	6258	0.5%
2021	27.5	−11.0%	234.3	−3.1%	1476.9	−1.0%	6425	2.7%
2022	23.9	−12.9%	226.0	−3.5%	1521.9	3.1%	6751	5.1%
2023	21.5	−10.1%	223.0	−1.4%	1473.0	−3.2%	6724	−0.4%
20-year period	−173.7	−89.0%	−244.4	−52.3%	−375.7	−20.3%	2548	61.0%
Average annual rate	−10.9%		−3.8%		−1.1%		2.6%

illustrates key prevailing trends between 2004 and 2023, showing a steady decline in the number of dairy farms, while milk production has remained relatively stable.

As for the number of farms, only 21.5 thousand farms were operating in 2023, compared to 195.2 thousand in 2004 [58]. This represents an average annual decline of 10.9%. During the 2004–2023 period, the number of dairy cows decreased by a factor of 2.2—from 467.4 thousand in 2004 to 223.0 thousand in 2023. Despite this decline, total milk production was only marginally affected, falling by approximately 20.3%, or 1.1% annually.

The number of farms decreased nearly ten times faster than annual milk output, illustrating the emergence of large-scale commercial dairy farms focused on business development. This marks a shift from keeping cows for family consumption toward market-oriented milk production. Between 2004 and 2023, the quantity of milk purchased for processing increased by 18%. In 2004, 59% of all agricultural holdings kept dairy cows, whereas by 2023 this share had dropped to just 14%.

Average milk production per farm increased at varying rates throughout the period, with an annual growth rate of 11.1%. The expansion of large-scale dairy farms improved their ability to invest in technology and innovation, leading to increased productivity. To better utilize investments in buildings and equipment, farms needed to scale up their production, resulting in a shift toward intensive dairy farming. Grazing systems were replaced by housing systems, which allowed for a significant increase in cow productivity—milk yield per cow rose by 61%, from 4176 kg in 2004 to 6724 kg in 2023.

The intensification and concentration of production raise concerns about the preservation of rural landscapes and the natural environment. The decline in the number of dairy farms, along with their withdrawal from more fertile land—where farmers have better opportunities to shift toward crop production—leads to a reduction in grassland and pasture areas and contributes to biodiversity loss. In addition, these changes are further reinforced by the use of intensive milk production technologies in large-scale dairy farms. Another concern is the issue of animal welfare. Keeping cattle indoors with limited space for movement or no access to outdoor areas results in animal health problems (such as hoof diseases and lameness) and increases consumer dissatisfaction regarding housing conditions.

4.2. Technical Efficiency of Lithuanian Dairy Farms

The data used to assess the technical efficiency (TE) of Lithuanian dairy farms were obtained from specialized dairy farms (farm type TF45; specialist milk) included in the Farm Accountancy Data Network (FADN). The dataset contained information from 913 observations covering the period 2017–2019. The period for the analysis was chosen from 2017 to 2019, because of the relatively stable period of activity between two crisis periods (after the price collapse starting in 2014 and the abolition of EU quotas in 2015 to the COVID-19 pandemic in 2020, and the war in Ukraine caused by Russia in 2022). To improve data reliability, the observations were combined into a balanced panel dataset of 176 farms over three years (2017–2019), calculating the average values of certain variables over that period. Typically, in studies that do not seek to analyze the rate of variability, a 3-year period is sufficient to remove the random effect of variability due to natural or other events. Since, as shown in Table 2, structural changes in farms are ongoing, combining data over a longer period would reduce the influence of changing technologies when assessing technical efficiency.

To classify farms into two clusters based on their production technologies, the statistical k-means clustering method was applied. Farms were grouped using two variables: livestock density (LU/ha) and self-sufficiency in home-grown feed (see

Table 3. Descriptive statistics of the data.

	All Farms		Cluster1-Extensive Farms		Cluster2-Intensive Farms
	Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.
Farms, number	176		117		60
Total farm output (EUR) y	153,262	299,152	73,222	84,589	308,007	463,193
UAA (ha) X1	107	115	79	75	161	155
Total farm labor (h) X2	6318	6091	4761	2484	9327	9167
Herd size (LU) X3	78.1	102.7	43.5	42.8	145.1	144.3
Intermediate consumption (EUR) X4	96,277	194,857	46,041	53,183	193,400	304,226
Capital (excluding herd and land) (EUR) X5	156,452	261,412	82,798	130,278	298,850	371,920
Livestock density (LU/UAA)	0.66	0.30	0.53	0.20	0.92	0.29
Share of home-grown feed	0.75	0.17	0.85	0.10	0.57	0.12
UAA productivity (milk/ha), t	2.49	1.47	1.81	0.88	3.81	1.48
Milk yield (kg per dairy cow)	5487	1553	5092	1363	6250	1624
Average cows per farm	52	72	28	28	97	104
Standard Production (EUR)	117,469	169,734	68,169	70,159	212,782	248,677
Total farm output per hectare (EUR)	1118	629	838	376	1659	669
Total farm output per LU (EUR)	1691	546	1627	536	1813	549
Total farm output per AWU (EUR)	39,531	33,059	28,814	21,452	60,251	41,015
Ratio of dairy output to total farm output	0.59	0.12	0.55	0.12	0.67	0.08
Operational subsidies, EUR	28,625	28,267	20,732	18,037	43,885	37,101
Operational subsidies per UAA, EUR	284	69	277	73	299	59
Operational subsidies per LU, EUR	532	331	629	367	344	90
Interm. consumption per LU (EUR)	1108	340	1090	342	1141	337
Interm. consumption as share of total output	0.67	67	0.69	0.18	0.64	0.09
Share of hired labor in total labor	0.28	0.16	0.18	0.27	0.49	0.34
Milk price (EUR/t)	269	52	257	53	293	40
Net value added (EUR)	55,030	89,686	30,700	35,503	102,067	134,118
Net value added per AWU (EUR)	15,060	12,767	11,900	11,328	21,170	13,256

).

The statistical analysis confirmed that there were significant differences between the clusters. T-tests showed statistically significant differences across both variables (p < 0.0001), confirming the validity of the clustering. Farms in the first cluster were characterized by lower intensity and higher self-sufficiency—they relied less on purchased feed and external inputs. The average livestock density in this cluster was 0.53 LU/ha, milk production per hectare was 1.81 tons, milk yield per cow was 5092 kg and self-sufficiency in feed production reached 85% (see Table 3). It is also worth noting that a high level of feed self-sufficiency indicates that these farms carry out more on-farm activities (such as sowing, forage harvesting, silage production, etc.), which explains why the scale of investment in extensive farms is not much lower than in intensive farms.

In contrast, farms in the second cluster were more intensive, with a higher livestock density (0.92 LU/ha), significantly higher milk productivity (3.81 milk t/ha, 6250 kg milk per cow), and lower self-sufficiency in feed (57%). These differences reflect distinct production technologies—extensive and intensive; therefore, in this study, the two clusters are referred to as “extensive farms” and “intensive farms.”

Table 3 presents the number of farms in each cluster and their main characteristics. The results show that 66% of the farms in the dataset are classified within the extensive cluster. These farms are, on average, 2 times smaller in terms of UAA compared to intensive farms, and the average number of LU per farm is more than three times lower. In intensive farms, approximately 49% of the labor force consists of hired workers, whereas in extensive farms this share is only 18% (see Table 3).

When evaluating economic indicators, intensive farms generated 2.0 times more income per hectare, labor productivity is also significantly higher, with intensive farms producing on average 2.15 times more output per hour of work compared to extensive farms, and the net value added per AWU was 1.8 times higher compared to extensive farms. It is noteworthy that both farm types received a similar level of direct payments per hectare, with intensive farms receiving 7.7% more. This difference can be explained by the fact that most direct support schemes are based on the number of hectares managed.

The next stage of the analysis involves the assessment of TE.

Table 4. Technical efficiency (TE) of Lithuanian dairy farms assessed using DEA meta-frontier analysis.

	Mean	SD	Min	Max
TE_VRS group frontier
Extensive	0.818	0.140	0.557	1.000
Intensive	0.894	0.101	0.653	1.000
TE_CRS group frontier
Extensive	0.718	0.184	0.316	1.000
Intensive	0.847	0.117	0.614	1.000
SE_group
Extensive	0.892	0.139	0.316	1.000
Intensive	0.950	0.069	0.734	1.000
TE_VRS meta-frontier
Extensive	0.777	0.135	0.519	1.000
Intensive	0.860	0.116	0.624	1.000
TGR
Extensive	0.950	0.049	0.806	1.000
Intensive	0.962	0.066	0.650	1.000

presents a summary of the efficiency analysis results. Intensive farms exhibited higher average TE compared to extensive farms, both relative to the group frontier and the meta-frontier.

The average TE of intensive farms was 0.894 when evaluated against their group frontier, and 0.860 relative to the meta-frontier. These figures suggest that, on average, intensive farms could reduce input use by approximately 10–14% while maintaining the same output level, depending on the chosen reference frontier.

By contrast, extensive farms demonstrated considerably lower efficiency scores—0.818 with respect to the group frontier and 0.777 against the meta-frontier. This implies that farms in this group could theoretically improve performance by reducing input use by 18–22% without compromising output. Moreover, extensive farms showed greater variability in efficiency (SD = 0.140), indicating higher internal heterogeneity. Both clusters included farms that achieved full efficiency (TE = 1.0), with approximately 25% of farms in the extensive group and 30% of farms in the intensive group reaching this benchmark. However, the minimum efficiency observed among extensive farms was significantly lower, at 0.557.

The disparity in TE is even more pronounced under the constant returns to scale (CRS) assumption; extensive farms were, on average, only 71.8% efficient, compared to 84.7% for intensive farms. The lowest CRS-based efficiency score among extensive farms was 0.316, indicating very poor resource utilization in some cases. These findings highlight efficiency losses associated with suboptimal production scale as captured by the CRS model.

The scale efficiency (SE) of intensive farms was higher—0.950 compared to 0.892 for extensive farms. This indicates that intensive farms are closer to the optimal scale of production and experience lower efficiency losses due to suboptimal scale. Extensive farms may benefit from reviewing their production structure to improve scale alignment.

Differences in TE across clusters (see Figure 1) indicate that intensive farms are characterized by a higher level of technological advancement, greater internal consistency, and more effective scale utilization. This is evidenced by the fact that a smaller proportion of farms in this group significantly lag behind the group-specific frontier, suggesting more efficient resource management.

In contrast, extensive farms are characterized by greater internal heterogeneity, which is reflected in a wider range of TE scores and a lower average value, as well as a larger proportion of farms operating further from the efficiency frontier. This heterogeneity may be influenced by various factors, including differences in agroecological conditions (e.g., soil quality, topography), access to advisory services, farm succession dynamics, differences in farmers’ experience and innovation uptake, access to financial resources, and farmers’ decisions on the continuation of agricultural activities. At the same time, this means that there is greater potential for improving performance within this group. Extensive farming systems, which rely more heavily on internal resources, tend to be more resilient under adverse external conditions due to their lower dependence on external inputs such as hired labor or purchased feed. Nevertheless, a comparative cost analysis conducted by Alvarez et al. (2008) [33] found that intensive dairy farms operate with lower average costs than extensive ones. Consequently, intensive farms maintain a competitive advantage in the long term.

The Technology Gap Ratio (TGR) quantifies the extent to which a group’s technological frontier falls short of the common meta-frontier. The TGR for intensive farms is 0.962, whereas for extensive farms it is 0.950. These values suggest that the technologies adopted by intensive farms are close to the most advanced available, while extensive farms remain further behind. This gap may indicate greater potential for technological improvement, or alternatively, reflect structural limitations inherent in the extensive production model.

To verify whether the differences in efficiency indicators between farm groups are statistically significant, we applied the non-parametric Mann–Whitney U test for mean comparisons. The results revealed statistically significant differences: TE (VRS) p = 0.0029, SE p = 0.0339, TGR p = 0.0016. These findings support the assumption that technological advancement and higher production intensity enable farms to utilize available resources more effectively and operate closer to the best-practice production frontier.

Correlation analysis showed that TE (VRS) among extensive farms was only weakly related to the SE and the TGR. This confirms the high internal diversity among farms and indicates that the level of efficiency in this group is influenced by various or imperceptible factors. In contrast, intensive farms showed a moderate positive correlation between SE and TGR (r = 0.65), which indicates that farms that make better use of economies of scale are also closer to the meta-frontier and more technologically advanced intensive farms make better use of economies of scale.

Efficient and inefficient producers were analyzed by separately comparing the key production characteristics of both groups: extensive and intensive.

Table 5. Characteristics of extensive farms grouped by technical efficiency under VRS.

	Efficient (TE (VRS) ≥ 0.9)		Inefficient (TE (VRS) < 0.9)		Significance 1
	Mean	S.E. 2	Mean	S.E. 2
Technical efficiency (TE (VRS)	0.985	0.00	0.738	0.01	***
Scale efficiency (SE)	0.930	0.02	0.870	0.02	*
UAA (ha)	103.6	15.7	65.24	5.48	*
Average cows per farm	41	5.62	21	2.08	**
Milk yield (kg per dairy cow)	5730	0.20	4716	0.14	***
UAA productivity (milk t/ha)	2.35	0.15	1.49	0.08	***
Livestock density (LU/UAA)	0.58	0.03	0.50	0.02	*
Share of home-grown feed	0.84	0.01	0.85	0.01	ns
Share of dairy cows in total LU	0.72	0.02	0.67	0.02	*
Capital Assets per UAA (EUR)	2420	211.8	1753	107.3	**
Number of LU per AWU	22.1	2.11	15.2	1.08	**
Share of hired labor in total labor	0.21	0.04	0.16	0.03	ns
Interm. consumption per LU (EUR)	1162	46.8	1047	41.7	ns
Ratio of dairy output to total farm output	0.61	0.01	0.51	0.01	***

^1,*—p-value < 0.05, **—p-value < 0.01, ***—p-value < 0.001, ns—difference not significant; ²—standard error.

presents the main characteristics and TE scores of extensive farms, which were divided into two groups based on their TE estimates. The first group—efficient farms (n = 43)—includes farms with a TE (VRS) score equal to or greater than 0.9, while the second group—inefficient farms (n = 73)—includes those with a TE score below 0.9. The mean TE (VRS) score was 0.738 for the technically inefficient producers, compared to a mean score of 0.985 for the efficient producers. The mean scale efficiency (SE) score was 0.870 for the inefficient producers and 0.930 for the efficient ones. The TE differences between the groups are statistically significant (p < 0.001), indicating that efficient farms operate closer to the frontier and make better use of scale economies.

The analysis revealed that efficient farms are larger both in terms of land area and the number of cows. Milk yield per cow and milk produced per hectare were 22% and 58% higher, respectively (p < 0.001), in efficient farms compared to inefficient ones. Efficient producers also demonstrated more effective use of labor, with a higher number of LU per AWU—22 LU per AWU in efficient farms vs. 15.2 LU per AWU in inefficient farms (p < 0.01). They also had significantly higher capital intensity (EUR 2420/ha vs. EUR 1753/ha; p < 0.01). These results suggest that efficient farms not only invest more in capital and technology but also utilize these resources more effectively. Moreover, they display a higher level of specialization, as indicated by the share of cows in total LU and the share of dairy output in total farm output.

For other tested variables—such as the share of homegrown feed in total feed, the share of hired labor in total labor input, and intermediate consumption per LU—no statistically significant differences were observed between the two groups.

The analysis revealed that technically efficient extensive farms are characterized by higher cow productivity, better scale effects, greater capital utilization, and more efficient use of labor resources. This suggests that efficiency is achieved by expanding the scale of production—both in terms of cow productivity and herd size—through more targeted investments and the ability to apply available technologies.

Therefore, the value of these results lies not only in confirming known efficiency patterns but also in uncovering the diversity within the extensive farming model, which is often treated as a homogeneous category. These findings support the need for more tailored policy instruments and suggest that future research could benefit from applying conditional DEA or meta-frontier models that take into account environmental and structural constraints in performance evaluation.

Table 6. Characteristics of intensive farms grouped by technical efficiency under VRS.

	Efficient (TE (VRS) ≥ 0.9)		Inefficient (TE (VRS) < 0.9)		Significance
	Mean	S.E.	Mean	S.E.
Technical efficiency (TE (VRS)	0.9775	0.01	0.8049	0.01	***
Scale efficiency (SE)	0.9538	0.01	0.9462	0.01	ns
UAA (ha)	171.6	28.0	148.08	13.4	ns
Average cows per farm	111.2	19.7	80.49	7.0	ns
Milk yield (kg per dairy cow)	6733	0.28	5660	0.13	**
UAA productivity (milk t/ha)	4.25	0.24	3.27	0.14	**
Livestock density (LU/UAA)	0.93	0.04	0.92	0.04	ns
Share of home-grown feed	0.55	0.02	0.60	0.01	ns
Share of dairy cows in total LU	0.69	0.01	0.64	0.01	*
Capital Assets per UAA (EUR)	2692	235	2730	145	ns
Number of LU per AWU	33.2	3.42	32.1	1.84	ns
Share of hired labor in total labor	0.47	0.05	0.50	0.04	ns
Interm. consumption per LU (EUR)	1192	57.9	1080	31.6	ns
Ratio of dairy output to total farm output	0.70	0.01	0.65	0.01	*

*—p-value < 0.05, **—p-value < 0.01, ***—p-value < 0.001, ns—difference not significant; S.E.—standard error.

presents a comparison of farms applying intensive production technologies, grouped by their TE level. The efficient group (n = 33, TE (VRS) ≥ 0.9) had an average TE (VRS) score of 0.9775, while the inefficient group (n = 27, TE (VRS) < 0.9) scored 0.8049. The mean scale efficiency (SE) score was 0.9462 for inefficient farms and 0.9538 for efficient farms. The difference in SE scores was not statistically significant, indicating that efficiency differences are more likely related to internal farm management rather than scale effects.

The analysis showed that efficient farms are larger in terms of both utilized agricultural area and number of cows—by 16% and 38%, respectively. Statistically significant differences (p < 0.01) were found for milk yield per cow and milk production per hectare, which were 19% and 30% higher, respectively, in efficient farms. Efficient farms also demonstrated a higher level of specialization (share of dairy cows in total LU and share of dairy output in total output) (p < 0.05).

It is noteworthy that the inefficient group had a slightly higher capital value per hectare (EUR 2730/ha) compared to the efficient group (EUR 2692/ha), although this difference was not statistically significant (p = 0.917). This indicates that capital intensity alone is not a determining factor of technical efficiency. It may suggest that inefficient farms are over-invested or that their investments are suboptimal, failing to yield proportional returns in terms of productivity or efficiency.

No statistically significant differences were observed between groups in terms of livestock density, capital and labor inputs, the share of home-grown feed, the share of hired labor, or intermediate consumption per livestock unit.

The results of the analysis suggest that productivity (milk yield per cow and milk produced per hectare) is the key factor determining the efficiency of intensive farms. Technical efficiency depends more on the management of available resources than on their absolute quantity.

A technical comparison of efficient and inefficient farms within the intensive group (Table 6) shows that most variables between the two groups do not differ significantly. These results confirm greater internal homogeneity among intensive farms, particularly in terms of capital intensity, herd size, and cost utilization, in contrast to the extensive group of farms. Thus, although the absence of significant differences may seem insignificant at first glance, it confirms a broader interpretation of how efficiency works differently in different production systems. This contrast reinforces the need to develop tailored efficiency-enhancing strategies, taking into account the specific circumstances and differences of each group.

Our findings are in line with other previous studies demonstrating the greater productivity and technical efficiency of intensive farms. The results of these studies can be interpreted as an economic justification of the trend towards intensification of dairy farms [3]. For example, Alvarez & Del Corral (2010) [35] found that both productivity and technical efficiency were higher in intensive farms compared to extensive ones in Spain; the technical efficiency of intensive farms was 0.971, while for extensive farms it was 0.931.

Latruffe et al. (2023) [20], using a latent class stochastic frontier model combined with a novel nested meta-frontier approach, examined dairy farms in France, Austria, and Ireland and found that, in all countries, intensive farms had higher technical efficiency. The study concluded that the largest productivity differences arose from the technologies used rather than inefficiency.

Similarly, Garcia-Covarrubias et al. (2023) [38] found that Ireland’s larger, more intensive farms operate closer to the production frontier (i.e., the technical efficiency score for intensive farms was 91%, compared to 82% for the other group).

Dakpo et al. (2022) [21], analyzing dairy farms participating in agri-environmental schemes using a latent class stochastic frontier model for each group, found that in both sub-samples, intensive farms were more efficient.

When comparing the performance of dairy farms across EU countries, the results indicated that intensive farms are more productive. Findings from Poczta et al. (2020) [13], who classified EU dairy farms into five clusters, confirmed that the most productive farms are medium and large operations, characterized by high specialization, large production scale, and an intensive production process. Likewise, Wilczyński et al. (2020) [53] showed that larger dairy farms (in terms of herd size) were the most efficient compared to smaller farms.

Efficiency is a key determinant of the long-term viability of dairy farms [30]. Given that intensive technology is more productive and intensive farms are more efficient, it can be concluded that the trend toward intensification is likely to continue.

5. Conclusions

The aim of this study is to assess the technical efficiency (TE) of dairy farms by classifying them based on the intensity of their production technologies. To account for environmental pressures—such as land use intensity, nutrient loads, pollution risks, and grassland degradation—farms were clustered using indicators identified in the literature, namely stocking rates and the share of on-farm produced forage, as proxies for the extensive or intensive nature of production.

This study contributes to the ongoing discussion on the economic advantages and limitations of intensive and extensive production systems. The findings provide empirical evidence on the main factors influencing milk production efficiency.

The results confirm that farms grouped according to their technological orientation exhibit significant differences in both production and economic outcomes. Extensive dairy farming is characterized by milk production that primarily relies on on-farm produced forage and low stocking rates. These farms are generally smaller in terms of land area and herd size, more reliant on internal rather than hired labor and less productive. Their average TE scores were 0.818 (group frontier) and 0.777 (meta-frontier), compared to 0.894 and 0.860 for intensive farms. Intensive farms exhibited higher scale efficiency (0.950 vs. 0.892) and a higher technology gap ratio (TGR) of 0.962 versus 0.950. These results suggest that the weaker performance of extensive farms is driven by both technological choice and technical inefficiency.

While extensive farms demonstrate a broader potential for performance improvement, this finding should be interpreted with caution. Our study revealed that, unlike intensive farms, extensive farms exhibit a high degree of internal heterogeneity. This underscores the need for further research to more precisely identify subgroups within this farming model. Such differentiation would allow for a more accurate evaluation of their economic performance and better inform targeted policy interventions. Although the meta-frontier framework accounts for technological heterogeneity between groups, it does not incorporate farm-specific constraints within groups. Future studies could therefore benefit from applying conditional DEA models or incorporating environmental variables to more accurately capture efficiency potential under such constraints.

Given the higher productivity and efficiency of intensive systems, the trend towards intensification is likely to persist. The analysis of farms showed that among efficient and inefficient farms in both technology groups, the most statistically significant indicators were those directly related to increased production volumes, i.e., milk productivity per cow, milk production per UAA, and a higher level of specialization. This means that farms pursuing long-term milk production strategies are likely to continue expanding their production.

In pursuit of improved environmental outcomes and higher animal welfare standards, the EU agricultural policy encourages the adoption of more extensive farming practices. However, in Lithuania—as in many other EU countries—the dairy sector is evolving in the opposite direction, moving toward more intensive production systems. This creates a fundamental policy dilemma: how to ensure the economic viability of dairy farms and strengthen the competitiveness of European dairy production in global markets, while simultaneously advancing environmental sustainability. The findings of this study indicate that intensive dairy farms are more productive than extensive ones. Consequently, farms aiming to enhance their economic viability are likely to continue shifting toward intensification.

These results underscore the need to align policy incentives with on-farm realities. Rather than relying on a uniform approach, policy instruments should be differentiated based on the structural and technological characteristics of farms. In particular, support for extensive systems must take into account their lower productivity potential and often greater vulnerability to market pressures.

More targeted and technology-specific research—particularly studies that segment farms by production system and investigate their motivations to expand or maintain existing production models—can provide valuable insights for policymakers. Such knowledge would support the design of better-aligned policy tools, capable of balancing economic performance with long-term sustainability objectives.