A combined multi-variate statistical analysis to establish dairy farm typologies in Cantabria

Computers and Electronics in Agriculture 221 (2024) 109007 Available online 10 May 2024 0168-1699/© 2024 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/bync-nd/4.0/). A combined multi-variate statistical analysis to establish dairy farm typologies in Cantabria Ib´ an V´ azquez-Gonz´ alez a , * , Elena García-Su´ arez b , Francisca Ruiz-Escudero b , Ana Isabel García- Arias a a University of Santiago de Compostela, Escola Polit´ ecnica Superior de Enxe˜ naría (Department of Applied Economics), Campus Universitario s/n, 27002 Lugo, Spain b Centre for Agricultural Research and Training (CIFA), Government of Cantabria, 39600 Muriedas-Cantabria, Spain ARTICLE INFO Keywords: Dairy farm categorisation Principal component factor analysis Hierarchical cluster analysis Northern Spain Complex sample ABSTRACT In the last few decades, dairy farms have undergone an intense process of structural adjustment. Despite this, dairy farming remains the most important agricultural activity in Cantabria (northern Spain), with many different types of dairy farm existing. However, there are few studies that have characterised and established typologies to understand this diversity. This study aimed to develop a method for characterising and categorising all the dairy farms in Cantabria (farm population) from a productive, economic and social point of view using combined multi-variate analysis techniques, including principal component factor analysis (PCFA) and hierarchical cluster analysis (HCA). For this purpose, 86 surveys were conducted on dairy cattle farms in Cantabria from 2016 to 2017 using stratified random sampling optimised with Neyman’s minimum variance allocation. The results, which relate to all the dairy farms in Cantabria (SPSS complex sample module), have enabled us to characterise and categorise these farms. The sector is mostly made up of farms with low-production levels. Their main characteristics are the importance of rented land, the use of pasture, good production management, a notable absence of young owners, a strong family link and moderate economic viability. The PCFA synthesised 22 production and socio-economic variables into 8 factors that reproduce 77.7 % of variance, half of these factors being economic in nature. The HCA, which used a double decision criterion to define the optimum number of clusters, has classified the dairy farms in Cantabria into four production typologies, which differ in the level of agricultural activity, profitability and diversification. In addition, a fifth group of farms was identified as singular cases. 1. Introduction The classification of farms into typologies is a common practice that has become widespread in recent decades (Ferreira-Golpe et al., 2021). It enables a series of individuals to be placed into homogeneous groups based on a set of variables and criteria (Schwering et al., 2022). Several classification criteria exist, and the most common are related to the production system and other socio-economic aspects (Sraïri and Lyoubi, 2003). The main advantages associated with the categorisation of agricultural production systems are that it reveals their structure, functioning and strengths and enables the identification of those aspects of production that need to be improved (Mądry et al., 2013). Furthermore, the establishment of typologies minimises the difficulties in understanding production systems and provides knowledge regarding their differences (Cortez-Arriola et al., 2015; Hassall et al., 2023). It is also useful for defining improvement strategies in the framework of future policy plans (Graskemper et al., 2021), promoting sustainable development (Castel Genís et al., 2010) and advising on sustainability (Andersen et al., 2007; B´ ankuti et al., 2020). Most of the existing classifications have employed multi-variate statistical analysis techniques, which enable variables of different natures to be used together; Ferreira-Golpe et al. (2021) used cluster analysis with three productive and two socio-economic variables to characterise the honey production sector in northwest Spain. Different techniques and methods exist, which vary depending on the objectives pursued and the nature of the data (Andersen et al., 2007). For example, Connell et al. (2007) used descriptive statistics for a technicalproductive characterisation of cattle systems in Anzoategui State (Venezuela). Silva et al. (2007) adopted frequency distribution to detect critical nodes of management processes in dual-purpose cattle * Corresponding author. E-mail address: [email protected] (I. V´ azquez-Gonz´ alez). Contents lists available at ScienceDirect Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag https://doi.org/10.1016/j.compag.2024.109007 Received 19 October 2023; Received in revised form 25 April 2024; Accepted 1 May 2024 Computers and Electronics in Agriculture 221 (2024) 109007 2 production systems in Zulia State (Venezuela). Siegmund-Schultze and Rischkowsky (2001) used and compared three different methods (cluster analysis, logistic regression and correspondence analysis) in the identification of the socio-economic characteristics of urban sheep keeping in west Africa. In these cases, the determination of the typology is the result of a single analysis; however, the typology can also be determined in several steps, for example, Perrot (1990) used a directed iterative method to aggregate with the help of experts the farms into typologies. Others, however, established typologies of livestock farm by combining multi-variate statistical analyses, first a factor analysis and then a cluster analysis. The latter has been used for the studies conducted by Serrano-Martínez et al., (2004b), who identified homogeneous groups of cattle farms in Le´ on (Spain). Cortez-Arriola et al. (2015), who captured the diversity of family-based dairy farms in Michoacan (Mexico). Blanco-Penedo et al. (2019), who classified the diversity of organic dairy farms in four European countries. G¨ okdai et al. (2020), who classified and characterised dairy goat farms in Italy and Turkey. Ruiz et al. (2020), who characterised extensive livestock farming systems (cattle, goats and sheep) in a protected area of Spain (Sierra Nevada). Schwering et al. (2022), who studied the typologies and motivations of farm management information systems in Germany. Other studies, such as those by Mateus Silveira et al. (2022), who represented the diversity of smallholder dairy production systems in a Brazilian semi-arid region, and Riveiro-Vali˜ no et al. (2009), who validated the dairy farm types in Galicia (Spain), went further and incorporated a third discriminant-type analysis. Dairy farming is a strategic part of the Spanish agri-food sector owing to its economic and social significance, with an annual turnover of around 13 billion euros and creation of more than 60,000 direct jobs (L´ opez Iglesias and Lainez Andr´ es, 2022). In recent decades, cattle farms have undergone an intense process of structural adjustment, characterised by a sharp reduction in the number of farms and employees, alongside an increase in production and the degree of intensification (Riveiro-Vali˜ no et al., 2009; Gonz´ alez-Mejía et al., 2018). Milk production in Spain can be divided into two different areas (Flores-Calvete et al., 2017). The northern strip, known as the Cantabrian Coast (Galicia, Asturias, Cantabria and the Basque Country), is the most important in terms of production (accounting for 79 % of the dairy farms and 56 % of milk production) (MAPA, 2023). Smaller production levels (on average 400 tonnes per year) and a diet based on fresh and preserved forage produced on the farm, which includes, to a varying extent, maize silage, and concentrate characterise the Cantabrian Coast farms (Flores-Calvete et al., 2017). Meanwhile, the rest of the country features farms with higher production levels and irrigated cropping systems and, in some cases, a greater reliance on concentrates and other purchased forage crops. Our study focused on Cantabria, a small autonomous community located in the north of Spain, where bovines constitute the economic, social and regional mainstay of the agricultural sector (Calcedo, 2013; V´ azquez-Gonz´ alez et al., 2023). Furthermore, it occupies a prominent position in the dairy cattle sector of the country (Ruiz-Escudero et al., 2023). In 2022, it was the third most important region in Spain by number of milk producers (977 units, equivalent to 8.41 % of the total) and sixth by number of cows (54,324 units, equivalent to 6.81 % of the total) and milk production (408,130 tonnes, equivalent to 5.57 % of the total) (MAPA, 2023). As each agricultural production system is different and faces specific problems whose solutions may be unique (Mądry et al., 2013), it is necessary to capture the heterogeneity and diversity of farms before making any decisions (Mateus Silveira et al., 2022). In Cantabria, there is a wide variety of dairy farms in relation to their production, socioeconomic and management characteristics (García-Su´ arez, 2021). However, there are few studies that have characterised them, even fewer that have established typologies and none that focused on all the farms in the territory; for instance, Blanco-Penedo et al. (2019) classified the diversity of organic dairy farms in 14 European regions (one Cantabria); Calcedo (2013) analysed the evolution of the milk production sector in Cantabria during the quota period; Celorio et al. (2011) performed a farm structural characterisation of the Pasiega local bovine breed in Cantabria; Doltra et al. (2018) identified and characterised four forage management systems in 40 Cantabria dairy farms (grazing, zero grazing, conserved forages and maize silage); and Salcedo et al. (2022) calculated the hydric footprint of 53 dairy farms in the north of Spain according to six food typologies (organic, conventional grazing, manger grazing, grass silage, maize silage, grass and maize silage). The above classifications have the limitations of referring to an individual farm (Doltra et al., 2018; Salcedo et al., 2022), a larger geographic scope (Blanco-Penedo et al., 2019; Salcedo et al., 2022) or a specific species (Celorio et al., 2011) and do not use multi-variate statistical analysis to establish typologies or the typologies established are according to feed/ forage management (Doltra et al., 2018; Salcedo et al., 2022). Due to the lack of knowledge presented above, this study aimed to: first, develop a method that characterises and categorises all the dairy farms in a given territory from a productive, economic and social point of view using combined multi-variate analysis techniques; second, apply this method to Cantabria to complete the available knowledge about its dairy sector. For this purpose, this study had three specific objectives: to obtain factors representative of variables of different nature (factorial characterisation), to obtain a representative productive and socioeconomic typology (farm typologies) and to characterise all dairy farms in Cantabria and the resulting typologies. We present results that refer to all dairy farms in Cantabria, not just those included in the surveys, using an unusual and effective method for that purpose (SPSS complex sample). In addition, we employed combined multi-variate statistical analysis to identify dairy farm typologies (non-existent in this area), considering the decision on the optimal number of groups in the cluster analysis using a twofold criterion. The obtained results are expected to expand the use of this methodological approach, provide further understanding of the diversity of the existing systems and help policymakers in formulating policies. It could also be used to establish links with livestock farm strategies. 2. Materials and methods The information used was obtained from a survey of dairy cattle farms in Cantabria conducted as follows. 2.1. Surveys A representative sample (surveys) of dairy farms in Cantabria was determined based on five strata of milk production in the 2015–2016 season 1 . The milk production strata were selected according to the population size distribution (Table 1) and the usual strata used in other Table 1 Sampling information (population size, number of surveys and elevation factor), based on milk production strata (tonne) in the 2015–2016 season. Milk production strata 2015–2016 (tonne) Population size (N) Number of surveys (n) Elevation factor (N/n) <100 364 8 45.5 [100;250] 482 17 28.4 [250;500] 325 19 17.1 [500;1000] 164 15 10.9 ≥1000 57 27 2.1 TOTAL 1392 86 16.2 Source: own elaboration. 1 The 1392 farm data used for sampling (monthly milk production) came from the mandatory cow’s milk declarations, provided by the regional administration, with continuous deliveries in the period from 1 April 2015 to 31 March 2016. I. V´ azquez-Gonz´ alez et al. Computers and Electronics in Agriculture 221 (2024) 109007 3 works (Flores-Calvete et al., 2017; Doltra et al., 2018). The sampling method employed was stratified random sampling optimised using Neyman’s minimum variance allocation for a sampling error of 5 % and confidence level of 95 % (V´ azquez-Gonz´ alez et al., 2023). The sample size (n) was given by the Equation 1, which includes the following parameters: Nh the population size, Sh the standard deviation, E the maximum sampling error (5 %), K the coefficient associated (1.96) with confidence level of 95 %, Y the population value of milk production in the 2015–2016 season (388,465 tonnes), Sh 2 the variance. The distribution by strata (nh) was given by the Equation 2, that uses any of the above parameters. Sample size (n) n=∑L h=1(Nh x Sh)2 E2x Y2 K2+∑L h=1Nh x Sh2(1) Distribution by strata (nh) nh =n x Nh x Sh ∑L h=1Nh x Sh (2) n =sample size. nh =sample size by stratum. Nh =population size in stratum h. E =maximum sampling error. Y =population value of variable i (Cantabria milk production in the 2015–2016 season). K =coefficient associated with confidence level. Sh =standard deviation of variable i (Cantabria milk production in the 2015–2016 season) in stratum j. Sh 2 =variance of variable i (Cantabria milk production in the 2015–2016 season) in stratum j. L =last stratum. The farms interviewed were selected through a random selection of cases function of the SPSS programme (v.21). This selection of farms was conducted separately for each of the five milk production strata and in two stages (initial and reserve farms), randomly and without replacement. The questionnaire, which was designed between September and October 2016, consisted of 623 items of information structured in six blocks: ownership, production, family structure, economic structure 2 , recent evolution and prospects (García-Su´ arez, 2021). Between November 2016 and February 2017, 86 surveys were conducted on dairy cattle farms in Cantabria through direct personal interviews lasting approximately 1 h. The farmers were previously contacted via telephone and re-contacted after the transcript for missing information. If one initial farm refused to be interviewed, it was replaced by a reserve belonging to the same strata and in a nearby area. Around half of the initial farms were replaced by reserves as they did not answer the telephone or did not agree to the interview for reasons of mistrust, fear and lack of time or interest. The questionnaire was mainly answered by the owners if they were the strategic decision-makers or the person who made such decisions. 2.2. Data analysis A double multi-variate analysis method was employed to classify the farms, first principal component factor analysis (PCFA) and second hierarchical cluster analysis (HCA). 2.2.1. Principal component factor analysis PCFA is a statistical dimension reduction method that represents a wide range of relationships between random variables through a subset of dimensions called factors (Vilela-Ferreira et al., 2021). The principal component method was employed to obtain the factors, and eigenvalues greater than 1 were selected as a criterion for the number of factors. To facilitate the interpretation of the factors, a rotation was performed using the Varimax method, whereas the adequacy of the sample was evaluated using Bartlett’s sphericity test and the Kaiser–Meyer–Olkin test (KMO). The stages carried out in the PCFA were as follows: A-Selection of variables 3 : a detailed selection was made based on the literature consulted and considering other issues: availability, quality and relevance (Kobrich et al., 2003; Mateus Silveira et al., 2022) (Table 2). B-Correlation study 4 : highly correlated variables were eliminated (R 2 ≥0,9) (Kobrich et al., 2003). C-Carrying out successive factor analyses until a valid one was found: the final valid analysis must be substantial (KMO >0.5 and significance <0.05); furthermore, all the variables must have high commonality (>0.5) and be interpreted in the rotated component matrix (correlation|r|>0.5). 2.2.2. Hierarchical cluster analysis The classification of the livestock farms into typologies has been performed using multi-variate HCA, the most common type of analysis, with the factor scores of the valid PCFA as the variables. Ward’s method was employed, and the dissimilarity measures used were the squared Euclidean distance (Caruso, 1997). A double decision criterion was considered to define the optimal number of clusters: a dendrogram and the calculation of the rates of variation of the clustering coefficients or ‘elbow rule’ (Graskemper et al., 2021; Mateus Silveira et al., 2022). The former, which is more commonly used, is graphical and is based on the interpretation of the dendrogram; the latter, which is analytical, is based on the calculation of the rates of change of clustering coefficients between successive stages (Kobrich et al., 2003; P´ erez and Santin, 2007). 2.2.3. Descriptive statistics and post-hoc tests The results section provides information on the descriptive statistics (mean values, coefficient of variation and confidence intervals at the 95 % level) for a total of 34 non-highly correlated production and socioeconomic variables. The values refer to the total population, to which the SPSS complex sample module has been applied. It is essential to use this module to build a sampling plan (csa plan file), with the specification of five milk production strata and their elevation factor, 5 using the sampling assistant for complex samples. Significant differences in the resulting typologies (Wald F-test) and multiple comparisons in the mean values of the groups are also determined via post-hoc tests 6 (5 % level) using the SPSS complex sample general linear model analysis (Zou et al., 2020). 2 The economic data collected in the survey correspond to 2016 and the rest of the variables to the time at which the survey was conducted. We are aware of the difficulty and limitations involved in collecting reliable economic information through a survey, so the results obtained are an approximation of the economic reality and should be interpreted with caution. 3 52 variables were selected (7 land base, 10 livestock, 9 milk production, 12 facilities and machinery, 4 family and work, 10 economic). 4 Bivariate Pearson’s or Spearman’s correlation was used, depending on the distribution of the data (Spearman’s if the distribution was not normal). A total of 18 highly correlated variables were eliminated. 5 Ratio for each stratum that measures the number of dairy farms in Cantabria (population size) divided by the number of surveys. 6 A post hoc test was calculated for significant differences or trends in the Wald F-test (p-value <0.1). I. V´ azquez-Gonz´ alez et al. Computers and Electronics in Agriculture 221 (2024) 109007 4 3. Results 3.1. Factor characterisation The final valid PCFA used 22 variables (2 milk production, 3 land base, 5 livestock, 4 facilities and machinery, 2 family and work, 6 economic) and generated 8 factors reproducing 77.7 % of the original variance. The analysis was satisfactory in statistical terms as the measure of adequacy (KMO) was high (0.708), Bartlett’s Sphericity test was significant (0.000) and the chi-squared value was equal to 95.1. Table 2 List of 52 variables initially selected for the PCFA, grouped according to categories (milk production, land base, livestock, facilities and machinery, family and work, economic). CATEGORY Variable Id. VARIABLES SIG. KS 1 Communality Ext. 5 MILK PRODUCTION 1 Milk production 15–16 per farm 2 0.007 2 Average milk price 15–16 ( € /L) 3,4 0.720 0.700 3 Milk quota purchase since 1992 (% of quota 15) 2 0.379 4 Milk production 15–16 per cow (L/cow) 2 0.728 5 Milk production 15–16 per UAA (L/ha UAA) 2 0.075 6 Milk production 15–16 per AWU (L/UTA) 2 0.423 7 Variation in milk production 16–17 (% of production 15–16) 3 0.000 8 Production offset 14–15 (% of quota 14–15) 3 0.030 9 Milk quota purchase since 1992 (kg) 3,4 0.000 0.787 LAND BASE 10 Average plot size (ha) 3 0.000 11 UAA maize (% of UAA total) 3,4 0.000 0.668 12 UAA grazing pastures dairy herd (% of UAA total) 3 0.000 13 UAA green-cut pasture manger dairy herd (% of UAA total) 3 0.092 14 UAA total (ha) 3,4 0.099 0.838 15 UAA used for pasture (ha) 2 0.023 16 Rented UAA (% of UAA total) 3,4 0.380 0.722 LIVESTOCK 17 Livestock units (LU) (milk +other cattle) 2 0.006 18 Production milk cows (% of total cows) 3,4 0.188 0.771 19 Rearing rate (% heifers ≥12 months of total cows) 3 0.288 20 Beef cows (% of total livestock units) 3,4 0.000 0.699 21 Stocking rate (LU/UAA) 3,4 0.826 0.793 22 Productivity per annual work unit (LU/AWU) 3 0.580 23 Total dairy cows (production +dry) 2 0.007 24 Average longevity before culling (lactations) 3,4 0.024 0.621 25 Cows monitored for milk yield (% of total cows) 3,4 0.000 0.751 26 Primiparous dairy cows (% of production cows) 3 0.306 FACILITIES AND MACHINERY 27 N◦of headlocks/N◦of beds 3 0.000 28 N◦of headlocks/dairy cow 2 0.248 29 N◦of beds/dairy cow 2 0.251 30 Slurry pit capacity per cow (m 3 /cow) 2 0.022 31 Year of construction most recent cowshed 3 0.010 32 Slurry pit capacity (m 3 ) 3,4 0.000 0.753 33 Most powerful tractor power per UAA (horsepower/ha) 3,4 0.087 0.812 34 Total surface area of cowshed (m 2 ) 2 0.000 35 Distance from cowshed to housing (m) 3,4 0.000 0.810 36 Maximum storage time slurry pit (months) 2 0.001 37 Most powerful tractor power (horsepower) 2 0.666 38 Cowshed surface area per cow (m 2 /cow) 3,4 0.075 0.857 FAMILY AND WORK 39 Number of family members 3 0.000 40 Age of the owner (years) 3 0.422 41 Total AWU 3,4 0.000 0.819 42 AWU paid labour (% of total AWU) 3,4 0.000 0.732 ECONOMIC 43 Total gross product (GP) (euros) 2 0.003 44 Net margin (euros) 2 0.000 45 Net margin per AWU (euros/AWU) 3,4 0.254 0.780 46 Net margin per 1000 L (euros/1000 L) 3,4 0.191 0.868 47 Net margin per UAA (euros/ha UAA) 2 0.155 48 Total cost (TC) (euros) 2 0.003 49 Milk revenues (% of GP) 3,4 0.001 0.805 50 Purchased food (% of TC) 3,4 0.816 0.914 51 General cost (% of TC) 3,4 0.080 0.771 52 External factor cost (% of TC) 3,4 0.030 0.824 1 Statistical significance of the Kolmogorov–Smirnov test (normal distribution contrast). If the p-value is equal or greater than 0.05 (normal distribution). 2 18 variables highly correlated (R2 ≥0.9), not analysed in PCFA. 3 34 variables not highly correlated included in the PCFA. 4 22 variables used in PCFA final valid analysis. 5 Extraction of communalities for 22 variables used in PCFA final valid analysis. Source: own elaboration. I. V´ azquez-Gonz´ alez et al. Computers and Electronics in Agriculture 221 (2024) 109007 5 Furthermore, all the variables used exhibited high communalities and high correlation coefficients in the rotated component matrix (see Tables 2 and 3). The factors have been defined according to the nature of the highly correlated variables (|r|>0.5) in the rotated component matrix (Table 3). Factor 1, called ‘production dimension and intensification’, explains 28.7 % of the total variance. It is positively correlated with the following variables: milk price, slurry pit capacity, total annual work unit (AWU) and percentage of paid labour, percentage of utilised agricultural area (UAA) under maize cultivation, stocking rate (total livestock units [LU]/ UAA) and amount of quota purchased since 1992. It is also negatively correlated with average cow longevity. Factor 2, called ‘economic profitability’, accounts for 11.4 % of the total variance. The best correlated variables, all of them being of an economic nature, are the economic profitability indicators, such as the net margin (NM) per AWU (NM/AWU) and the NM per volume of milk produced (NM/1000 L), both positively correlated. Factor 3, called ‘extensification’, explains 9.2 % of the total variance. The variable positively correlated with this factor is the total UAA and the variables negatively correlated with it are the stocking rate and the power in horsepower (hp) of the most powerful tractor on the farm (less than 10 years old) expressed as hectares (hp/ha). Factor 4, labelled ‘high-production-cost structure’, explains 6.8 % of the total variance. The variables positively correlated with this factor are the percentage of purchased food over total costs and the variable negatively correlated with it is the percentage of general costs over total costs. Factor 5, defined as ‘production specialisation’, explains 6.3 % of the total variance. The variables positively correlated with this factor are the percentage of UAA rented as well as the percentage of cows under a milk yield monitoring regime and cows producing milk. Factor 6, defined as ‘economic specialisation towards milk’, explains 5.5 % of the total variance. The variable positively correlated with it is the percentage of revenues from milk sales compared with gross product 7 and the variable negatively correlated with it is the percentage of beef cows over LU. Factor 7, defined as ‘external factors’, explains 5 % of the total variance. The most important variables, which are positively correlated with it, are the distance from the cowshed to the house and the percentage of external factor costs (land rent, paid labour and interest on loans) over total costs. Table 3 Interpretation of the factors resulting from final valid dairy cattle PCFA (explained variance, significance, identification of variables and correlation coefficient in the rotated components matrix). Factor→ Eigenvalue→ %Variance→ (accumulated) Name → Meaning factor Variables (Variable Id.) → name Correlation with factor 1 Production dimension and intensification→ Higher milk prices, dimension of the facilities, investments, labour needs, paid labour, maize cultivation and stocking rate and lower cow longevity (2) →Average milk prize 15–16 ( € /L) 0.525 (24) →Average longevity before culling (lactations) −0.570 (32) →Slurry pit capacity (m 3 ) 0.836 F1→6.32→ (41) →Total AWU 0.861 28.7 %→ (28.7 %) (42) →AWU paid labour (% of total AWU) 0.651 (11) →UAA maize (% of UAA total) 0.736 (21) →Stocking rate (LU/UAA) 0.518 (9) →Milk quota purchase since 1992 (kg) 0.785 F2→2.52→ 11.4 %→ (40.2 %) Economic profitability→ Higher farm income per annual work unit and per volume of milk (45) →Net margin per AWU (euros/ AWU) 0.800 (46) →Net margin per 1000L (euros/ 1000 L) 0.922 Extensification→ Higher utilised agricultural area and lower stocking rate and power per utilised agricultural area (14) →UAA total (ha) 0.747 F3→2.02→ (21) →Stocking rate (LU/UAA) −0.641 9.2 %→ (49.3 %) (33) →Most powerful tractor power per UAA (horsepower/ha) −0.820 F4→1.5→ 6.8 %→ High-production-cost structure→ Higher specific cost of purchased food and lower general cost (50) →Purchased food (% of TC) 0.929 (56.2 %) (51) →General cost (% of TC) −0.566 Production specialisation→ Greater relevance of rented land and cattle management (milk yield monitoring regime, production milk cows) (16) →Rented UAA (% of UAA total) 0.629 F5→1.38→ (25) →Cows monitored for milk yield (% of total cows) 0.653 6.3 %→ (62.5 %) (18) →Production milk cows (% of total cows) 0.649 F6→1.22→ Economic specialisation towards milk→ Higher revenues from milk sales and lower relevance of beef cows. (49) →Milk revenues (% of GP) 0.708 5.5 %→ (68.0 %) (20) →Beef cows (% of total livestock units) −0.803 F7→1.09→ 5.0 %→ (73.0 %) External factors→ Greater distance from cowshed to house and the percentage of external factor costs (land rent, paid labour and interest on loans) (35) →Distance from cowshed to housing (m) 0.861 (52) →External factor cost (% of TC) 0.650 F8→1.03→ 4.7 %→ (77.7 %) Animal welfare→More space for livestock (38) →Cowshed surface area per cow (m 2 /cow) 0.893 1 Only highly correlated variables (|r|>0.5) are presented. Source: own elaboration. 7 The following five revenues are considered: sale of milk, subsidies, cattle, other livestock and other agricultural revenues (processing of agricultural products, fertilizers, crops, insurance, etc.). I. V´ azquez-Gonz´ alez et al. Computers and Electronics in Agriculture 221 (2024) 109007 6 Finally, Factor 8, defined as ‘animal welfare’, explains 4.7 % of the total variance. The only single variable that is highly positively correlated with it is cowshed area per cow. 3.2. Farm typologies Out of the whole sample (86 surveys), we decided not to include three farms in the classification analysis as they always formed a different group in the different HCAs conducted due to the very different values (outliers) of some of the 22 variables used in the PCFA. This group of three farms, hereafter referred to as singular cases (SC), will be considered in the characterisation of the resulting typologies. 3.2.1. Determination of the number of clusters The dendrogram shows that the optimum number of clusters is 4, drawing a vertical line at a distance between 14 and 20 points. This interval gives the maximum distance for the horizontal lines separating the clusters for all the clusters with respect to the later (right) or earlier (left) stages (Fig. 1) (Kobrich et al., 2003). The graphical representation of the analytical solution, relative to the clustering rates at the different stages, suggests, as does the dendogram, that the optimal number of clusters is 4. It is noticeable how the slope increases at a lower number of clusters, e.g. 3 (Fig. 2). 3.2.2. Characterisation of dairy farms in Cantabria In relation to the land base, the holdings have an average of 28.6 ha of UAA, with relatively large plots (2.7 ha), where renting is the main tenancy regime (54.7 % of the UAA). The amount of land dedicated to fodder maize is low, accounting for less than 5 % of the UAA; contrarily, pasture under grazing is the main use, with about half of the UAA (Table 4). The average herd size per farm is 46 dairy cows, of which a high percentage is in production or being monitored for milk yield (82 % and 63 %, respectively). The average stocking rate is 2.9 LU/ha, the average longevity before becoming cull cows is 4.2 lactations and the rearing rate is 41.1 %, which might seem high for these herd longevity values. The diversification of production into other livestock activities, such as beef cows, is not widespread, accounting for less than 5 % of the total LU. In terms of milk production, the establishment of milk quotas in Spain in April 1992 and their subsequent abolition in March 2015 have conditioned the evolution of dairy farms. Since 1992, farms bought an average quota of 153,000 kg of milk. Furthermore, during the period immediately before (2014–2015 season) and after (2016–2017 season) the elimination of quotas, an opposite productive behaviour was detected. In the first period, milk production was almost 10 % lower than the available quota, whereas in the second period, it slightly Fig. 1. Dendogram. Fig. 2. Graphical representation of the variation rates of the clustering coefficients (elbow rule). Source: own elaboration I. V´ azquez-Gonz´ alez et al. Computers and Electronics in Agriculture 221 (2024) 109007 7 increased. In relation to the facilities and machinery available, it should be noted that on average, the most recent cowsheds are amortised, as they are more than 30 years old. These cowsheds are located at an average distance from the housing of 784 m and have other characteristics such as an average stall area per cow of 12.7 m 2 and 1.03 headlocks per bed. The mechanisation index, measured as the ratio of the power (hp) of the most powerful tractor on the farm (less than 10 years old) per hectare of UAA, is 4.6 hp/ha. Regarding the characteristics of the owner, family and work, almost Table 4 Production and socio-economic characterisation of the resulting clusters. Results elevated to the population as a whole, taking as a reference the 34 non-highly correlated variables used in the PCFA. GROUPS Desc. St. Complex Sample 4 GLM Variable Id. VARIABLES G1 G2 G3 G4 CS Total Coeff. Var. LCI UCI F Wald Sig 3 . N◦Surveys 19 42 15 7 3 86 Population (n◦farms) 5 67 559 488 222 56 1392 MILK PRODUCTION 7 Variation in milk production 16–17 (% of production 15–16) 1.88a 2 0.70b 1.48ab 2.58a 0.00c 1.3 0.432 0.18 2.42 2.87 0.028 2 Average milk price 15–16 ( € /1000L) 1 313a 295b 270c 259c 264c 280 0.012 279 287 10,8 0.000 8 Production offset 14–15 (% of quota 14–15) 15.7a 0.5ab −19.4bc −15.5b −31.6c −9.6 −0.304 −15 −3.8 4.42 0.030 9 Milk quota purchase since 1992 (1000 kg) 1 605a 213b 53.4c 34.1c 352ba 153 0.085 127 179 13.8 0.000 LAND BASE 14 UAA total (ha) 1 55.0a 35.8b 20.7c 19.7c 30.6bc 28.6 0.092 23.4 33.9 9.21 0.000 16 Rented UAA (% of UAA total) 1 68.8b 61.6b 49.4bc 37.5c 83.1a 54.7 0.086 45.3 64 5.84 0.000 11 UAA maize (% of UAA total) 1 35.4a 3.3b 4.0b 0.0c 10.3ab 4.8 0.208 2.8 6.8 7.89 0.000 12 UAA grazing pastures dairy herd (% of UAA total) 26.1b 34.3b 63.6a 59.4ab 56.2ab 49.1 0.092 40.1 58.1 3.04 0.022 13 UAA green-cut pasture for manger dairy herd (% of UAA total) 9.1 27.4 28.6 28.7 45.5 27.9 0.183 17.7 38.1 1.37 0.251 10 Average plot size (ha) 3.0ab 3.2a 2.7ab 1.8b 1.1b 2.7 0.058 1.9 3.5 2.19 0.077 LIVESTOCK 18 Production milk cows (% of total cows) 1 86.6a 85.4a 76.7b 84.4a 86.7a 82.3 0.012 80.3 84.3 4.34 0.003 26 Primiparous dairy cows (% of production cows) 36.9 31.4 31.1 39.8 39.6 33.2 0.073 28.4 38.1 1.0 0.410 25 Cows monitored for milk yield (% of total cows) 1 100a 100a 24.4b 33.3b 100a 62.9 0.099 50.5 75.2 26.7 0.000 19 Rearing rate (% heifers ≥12 months of total cows) 36.1 35.2 38.8 61.9 42.7 41.1 0.079 34.6 47.5 1.96 0.107 24 Average longevity before culling (lactations) 1 3.1b 4.3ab 4.8a 2.9b 4.6a 4.2 0.048 3.8 4.6 12.54 0.000 21 Stocking rate (LU/UAA) 1 3.6a 2.8b 2.6bc 3.6a 2.1c 2.9 0.056 2.6 3.2 5.25 0.001 20 Beef cows (% of total livestock units) 1 0.0b 2.2ab 1.9ab 16.8ab 6.8ab 4.5 0.365 1.2 7.7 3.24 0.016 22 Productivity per annual work unit (LU/AWU) 56.8a 35.3b 28.8b 45.4ab 36.9b 35.7 0.069 30.8 40.6 8.15 0.000 FACILITIES AND MACHINERY 31 Year of construction the most recent cowshed (over 1900) 99a 95a 74b 71b 90ab 84 0.002 77 91 3.99 0.005 35 Distance from cowshed to housing (1000 m) 1 0.97a 0.21b 0.092b 0.96ab 11.5a 0.784 0.459 0.07 1.49 2.43 0.055 32 Slurry pit capacity (100 m 3 ) 1 12.4a 3.69b 2.72bc 1.57c 4.71b 3.47 0.084 2.89 4.06 9.09 0.000 33 Most powerful tractor power per UAA (hp/ha) 1 2.9b 4.3ab 5.3a 4.2ab 5.1ab 4.6 0.093 3.75 5.46 3.18 0.018 38 Cowshed surface area per cow (m 2 /cow) 1 13.4a 14.1a 11.9a 7.5b 25.3a 12.7 0.086 10.5 14.9 7.6 0.000 27 N◦of headlocks/N◦of beds 1.00 1.06 1.03 1.00 0.98 1.03 0.024 0.98 1.08 1.21 0.314 FAMILY AND WORK 39 Number of family members 3.7 3.6 3.5 3.0 3.5 3.5 0.053 3.1 3.83 0.39 0.816 40 Age of the owner (years) 49.8 52.6 51.8 51.0 46.6 51.7 0.032 48.4 55.0 1.49 0.213 41 Total AWU 1 3.6a 2.3b 1.8bc 1.6c 1.8bc 2.1 0.046 1.86 2.23 4.61 0.002 42 AWU paid labour (% of total AWU) 1 37.9a 7.5b 1.0c 0.0 21.6ab 6.1 0.190 3.8 8.4 9.31 0.000 ECONOMIC 45 Net margin per AWU (1000 euros/AWU) 1 38.9a 9.6b 2.9b 12.0b 11.9b 9.1 0.197 5.55 12.7 3.94 0.006 46 Net margin per 1000 L 1 124 51 17 115 62 53 0.424 8.3 98.5 1.7 0.158 49 Milk revenues (% of GP) 1 83.5a 81.9a 71.4b 58.2c 70.8bc 74.1 0.023 70.7 77.4 8.82 0.000 50 Purchased food (% of TC) 1 41.0b 50.0a 38.7b 47.1ab 34.1b 44.7 0.039 41.3 48.1 5.68 0.000 51 General cost (% of TC) 1 20.7c 22.3bc 35.2a 28.9ab 24.7b 27.9 0.045 25.4 30.4 5.82 0.000 52 External factor cost (% of TC) 1 11.9b 6.2c 6.3c 8.2bc 19.9a 7.4 0.163 5.0 9.8 3.76 0.007 1 22 variables used in PCFA final valid analysis. 2 Subscript with a different letter represents the statistical significance between the groups (p-value <0.05). 3 Significance level in black (p-value <0.05). 4 Descriptive statistics: coefficient of variation (standard deviation/mean value) and confidence intervals at the 95 % level (lower LCI and upper UCI). 5 This variable represents the number of all dairy farms in the territory (Cantabria) that belongs to each group. Source: own elaboration. I. V´ azquez-Gonz´ alez et al. Computers and Electronics in Agriculture 221 (2024) 109007 8 all the dairy farms in Cantabria are family farms (98.6 %), mainly with the legal status of the owner as a natural person (58.9 %), followed by single-family companies (21.7 %), where all the partners are family members living in the same household and multi-family companies (18 %) (García-Su´ arez, 2021). The average number 8 of family members is 3.5, with the average age of the owner being 51.7 years. The average number of AWU is 2.1 AWU, of which 6.1 % is paid labour. In relation to the economic information for 2016, the average NM per AWU is 9,153 € /AWU, and the NM per unit of milk produced is 0.053 € /L. In terms of revenues, milk is by far the main income source (74.1 %), and in terms of expenditure, purchased feed is the main cost (44.7 %). Finally, the production structure of the sector is mostly made up of farms with low production, with 60.7 % of dairy farms producing less than 250 tonnes (Table 5). 3.2.3. Characterisation of farm typology Of the 34 variables analysed in Table 4, 9 do not substantially differ in mean values between typologies (p-value >0.05); in these cases, the mean value of most of the groups is within the confidence intervals (LCI and UCI). The variables, grouped in categories, are as follows: land base (percentage UAA devoted to green-cut pasture for in-manger feeding, average size of the plots), livestock (percentage of primiparous dairy cows, rearing rate), facilities and machinery (distance from the cowshed to the house, headlocks/bed ratio), family and work (number of family members, age of the owner) and economic (MN/1000 L). In the HCA, a total of four groups have been obtained, which are described below, together with the group of SC. Group 1–‘Farms with high production and profitability’. This group is made up of 19 of the farms surveyed, equivalent to 4.8 % of the dairy farms in Cantabria; all of them belong to the strata with the highest milk production (500 tonnes or more). Their land base is characterised by a higher UAA, a higher level of forage maize cultivation and rented land. Their livestock is characterised by a larger herd size, increased stocking rate, higher milk productivity per head, more controlled management (percentage of cows in production and under milk yield monitoring) and specialisation in milk production (absence of beef cows); however, the average longevity before culling is among the lowest. In terms of production, these farms, which have the highest milk production volume, received a higher price per litre of milk, have bought a larger amount of the milk quota and, in the last milking season (2014–2015) before the milk quota was abolished (March 2015), registered a considerable excess of production, which differs from the overall situation for all the farms in Cantabria. In relation to the facilities and machinery, their cowsheds were constructed more recently and are located at a greater distance from the house (significant trend); they also have a larger surface area per cow and a greater slurry pit capacity. The power ratio of the most powerful tractor per hectare of UAA has the lowest values, which may be due to the greater UAA available. Regarding labour, this group has higher labour requirements and a greater percentage of paid work (six times higher than the average for the sector). On the economic side, Group 1 has a greater economic profitability per worker (MN/AWU) as it combines higher milk productivity per AWU and unitary NM (NM/1000 L). Furthermore, a strong economic specialisation exists towards dairy, with this group having the highest percentage revenues from the sale of milk; however, general costs are lower due to the higher proportion of specific costs. Group 2–‘Farms with medium production and profitability’. This is the largest group, comprising 42 of the farms surveyed and representing 40.2 % of the dairy farms in Cantabria. Group 2 is made up of farms with lower production, all of them with a production of at least 100 tonnes; those with production between 250 and 500 tonnes (46 %) are the largest. The land base is characterised by an UAA that is substantially smaller than that of Group 1, and renting is the main land tenure regime, as in Group 1. However, the importance of forage maize is less, and the area under green-cut pasture or grazing is greater. As in Group 1, there is a high percentage of cows in production, all of them being monitored for milk yield; however, the longevity of the herd is higher, which is possibly associated with a lower stocking rate and milk productivity. The characteristics related to milk production (purchased quota, milk price and production variation), annual labour requirements and the number of paid workers are also lower than in group 1. The cowsheds are recently built and well-sized buildings, as in Group 1, but the slurry pit capacity is much lower. Group 2 has an NM/AWU that is much lower than that of Group 1 and similar to the average for the sector. Similar to Group 1, it specialises in milk production, as can be seen in the high dependence on milk revenues and the low percentage of general costs. However, the percentage of external factors is the lowest of all groups and indicates lower dependence on external resources (land, labour and capital). Group 3–‘Farms with low production and profitability’. Group 3 is made up of 15 of the farms surveyed, representing 35.1 % of the dairy farms in Cantabria. This group is composed almost entirely of farms producing less than 250 tonnes, of which the majority (56 %) produce less than 100 tonnes. Group 3 is characterised by a reduced surface, mainly dedicated to grazing. The herd is smaller in size and productivity (LU/AWU), as is the percentage of cows in production and under milk yield monitoring. Furthermore, the longevity of the cows is the highest, close to 5 lactations. The average price received for the milk sold, together with the negative evolution of production in the last quota campaign and the volume of quota purchased, has lower values different (p-value <0.05) from the two previous groups. The same is true for the age of the cowsheds. The power ratio of the most powerful tractor per hectare of UAA is the highest among all the groups due to the smaller size of the farms. Moreover, the annual labour requirements are lesser, and there is virtually no paid labour. In the economic side, Group 3 has the lowest profitability per worker, less than € 3,000 (MN/AWU), which shows the economic constraints they are experiencing and the need to supplement their incomes with other revenues. The percentage of expenditure on Table 5 Percentage distribution of dairy cattle typologies by milk production strata (tonne). Milk production strata 2015–2016 (tonne) GROUPS <100 [100;250] [250;500] [500;1000] ≥1000 Total G1. Farms with high production and profitability 0.0 0.0 0.0 49.3 50.7 100 G2. Farms with medium production and profitability 0.0 30.4 45.9 19.5 4.2 100 G3. Farms with low production and profitability 55.9 34.8 7.0 2.2 0.0 100 G4. Diversified farms with low production 41.1 51.2 7.7 0.0 0.0 100 G5. Singular cases 0.0 50.3 30.3 19.4 0.0 100 Total 26.1 34.6 23.3 11.8 4.1 100 Source: own elaboration. 8 In the case of multi-family companies and non-family companies, the number of members considered is the number of partners. I. V´ azquez-Gonz´ alez et al. Computers and Electronics in Agriculture 221 (2024) 109007 9 purchased feed is lower than the average, indicating the greater use of feed produced on the farm (greater autonomy), although the percentage of general costs is higher. Group 4–‘Diversified farms with low production’. This group is made up of 7 of the farms surveyed, representing 15.9 % of the dairy farms in Cantabria. It is composed of farms with low production; however, it has more farms with higher production than Group 3, since the majority (51 %) of these farms produce between 100 and 250 tonnes. Although Group 4 is similar to Group 3, some differences exist. The size, with 20 ha of UAA, is small, as is the case for Group 3; however the degree of renting is lower. In terms of livestock, despite having a similar livestock size, Group 4 has a higher percentage of cows in production and a higher stocking rate. Furthermore, the presence of beef cattle and a high rate of rearing seem to indicate a diversification of activity. The price received for the milk and the amount of quota purchased are the lowest, exhibiting no significant differences from those of Group 3. This also happens with the age of the herd, capacity of the slurry pit, annual labour requirement or number of paid workers. However, the available surface area per cow is substantially lower. Despite its small production size, Group 4 has a much higher NM, both per litre and per worker, than Group 3, and even higher than the average for dairy farms in Cantabria. It has the lowest percentage of revenues from milk sales among all the groups (58.2 %), which is compensated for by the money obtained from the diversification of agricultural activity (subsidies 22.3 %, calves 11.2 %, rearing heifers 4.2 %, cull cows 4.1 %). Group 5–‘SC’. This group is made up of three of the farms surveyed, representing 4 % of the dairy farms in Cantabria. The SC have a larger production structure than those of Groups 3 and 4; they are farms belonging to the intermediate-size strata (from 100 to 1000 tonnes), with half of them producing between 100 and 250 tonnes. Group 5 presents some unique production data, such as the high degree of renting (83 % of the UAA), small plots (1.1 ha), greater longevity before culling (4.6 lactations) and lower stocking rate (2.1 LU/ ha). In relation to milk production, they have purchased a substantial proportion of the quota since 1992, but the recent evolution has been marked by a much lower production than their potential; furthermore, the average price received for their milk is the lowest (264 € /1000 L). The most remarkable aspect in relation to the facilities is the greater distance of the cowshed from the house (11 km) and the greater capacity of the cowshed (25 m 2 /cow). On the economic side, their profitability is close to the average, but they have higher external factor costs (20 % of the total), which seems to be conditioned by the higher percentage of rented land, paid labour and investments made. 4. Discussion 4.1. PCFA Quantitative variables were used in the PCFA, which is the most common type of analysis (Mądry et al., 2013; V´ azquez-Gonz´ alez et al., 2022). The same authors, who were aware that the selection of variables depends on the production system analysed and the objective pursued in the characterisation, found that the most used technical variables were surface area, size and stocking rate, paid labour, feed supply and productivity; in terms of economic variables, the most used variables were revenues, expenditure and margins (income). Kaouche-Adjlane et al. (2015), in establishing a typology of dairy farms in Algeria, used variables related to ownership (age, education), structure (land base, livestock and equipment), management (feeding, production and reproduction) and economy. Maseda et al. (2004), in categorising family dairy farms in Galicia (northwest Spain), used 94 variables related to the location of the farm, family structure, sources of income, production, characteristics of the cowshed, characteristics of the facilities and transit routes. In our case, the production and socio-economic variables used that were related to milk production, land base, livestock, facilities and machinery, family and work and economy largely coincided with those used in the literature. The results obtained from the PCFA regarding the number of factors (8) and the percentage of variance (77.7 %) were consistent with those obtained in other studies characterising dairy farms with variance ranging from 50 % in England and Wales (Gonz´ alez-Mejía et al., 2018) and other European regions (Blanco-Penedo et al., 2019) to 84.2 % in Michoacan (Mexico) (Cortez-Arriola et al., 2015). The number of factors may vary from 3 (Kaouche-Adjlane et al., 2015; Gonz´ alez-Mejía et al., 2018) to 16, as reported by Maseda et al. (2004). The eight factors obtained have a decreasing importance in the percentage of variance explained, something that also occurs in the works consulted and is a characteristic of the analysis. In terms of the nature of the factors, the economic aspect is of greatest importance as it is present in four of the eight factors obtained (Factors 2, 4, 6 and 7). Thus, Factor 2, called economic profitability, is defined by two very important variables in the economic viability of a farm (NM/AWU and NM/1000 L). Bach et al. (2020) argued that in the economic analysis of a dairy farm, it is necessary to pay attention to milk production and to the unit margin, variables that were considered in our study. The Factor 4, referred to as the high-production-cost structure, is defined by a higher percentage for purchased feed costs; Salinas-Martínez et al. (2020) attributed higher feed costs to large farms owing to their greater dependence on concentrates. Factor 6, called economic specialisation, corresponds to farms whose income mainly depends on dairy farming. The European Commission applies a similar concept when defining the concept of technical economic specialisation, when at least 66 % of the gross margin is directly associated with this activity (EC, 2012). P´ erez- M´ endez et al. (2020), who investigated how health and reproduction affect the technical efficiency of dairy farms in Asturias (northwest Spain), have defined as specialised dairy farms where milk accounts for more than 90 % of sales revenues. Finally, Factor 7, referred to as external factor cost, is related to those farms that have fewer of their own resources (land, labour and capital). The number of factors obtained, their interpretation and the position they occupy exhibit similarities to those of other studies. We considered that the resulting factors can be grouped into two categories, basic and complementary, as reported by Serrano-Martínez et al., (2004a). The basic factors are ‘dimension and intensified production’ and ‘economic profitability’, which reproduces a higher percentage of the variance and is related to size, intensified production and economic profitability. The variables that define these factors, such as size, production, management, labour and economic factors, are considered to be necessary when defining production systems to characterise and classify farms (Cortez- Arriola et al., 2015; Gonz´ alez-Mejía et al., 2018). There are numerous complementary factors that reproduce a smaller percentage of the original variance (Serrano-Martínez et al., 2004b). These factors are defined by specific variables, for example, Factor 3, called extensification, is defined by three variables that take into account surface area (UAA, stocking rate and power/area ratio). Mateus Silveira et al. (2022) reported that surface area is a key variable when characterising extensive production systems. Factor 5, referred to as production specialisation, corresponds to farms with more controlled management. P´ erez-M´ endez et al. (2020) stated that all the specialised dairy farms in the study were monitoring milk yield. Finally, Factor 8, called animal welfare, is defined by a single variable, cowshed area per cow, which is considered in some studies to be a measure of animal welfare (García-P´ erez et al., 2022). 4.2. Identification and characterisation of farm typologies In our study, before the classification, we decided not to consider three farms (SC) in the HCA as they had very different values (outliers) in some of the 22 variables used in the PCFA. The same decision has been made in the previous literature; thus, Sraïri and Lyoubi (2003) and Cortez-Arriola et al. (2015) decided not to consider these farms in the classification analysis (2 and 1 farms respectively) and treated them as a I. V´ azquez-Gonz´ alez et al.