Classification variables of cattle farms in the mountains of León , Spain

This paper reports the f irst stage in the establishment of a farm classif ication system: the identif ication of classification variables. The aim of this work was to identify the variables most appropriate for characterising the cattle farms of the mountains of León, Spain, using multivariate and principal components analysis. The information used was obtained by surveying the cattle farms of the Montaña de Riaño (León) area from 1996-1998. The 35 variables taken into account in the principal components analysis refer to labour, the stock base, production, land use and farm economy. Seven principal components were found to explain 67.1% of the total variance.

In the Spanish province of León, the mountainous areas are of relatively high importance.The majority of the area's mountains are concentrated in two areas: Montaña de Riaño and Montaña de Luna.
The population density fell from 1981 to 2001 in both these areas: from 12.0 to 8.4 km -2 in Montaña de Riaño, and from 19.5 to 15.8 km -2 in Montaña de Luna (Junta de Castilla y León, 1992León, , 2002b)).At the same time, a number of environmental problems appeared (such as the proliferation of bushes, erosion, the loss of biodiversity and fires), frequently associated with changes in the use of the territory by agricultural and stock raising interests (Caraveli, 2000;MacDonald et al., 2000).An example of the quantitative importance of these changes is the change in the percentage of land classified as being of agricultural use.In the Montaña de Riaño area this fell by 38.1% between 1982-1989 while the Montaña de Luna area lost 14.6%, and by 32.1% and 17.3% respectively from 1990-1999 (Lavín, 1996;Junta de Castilla y León, 2002a).
At the same time, both areas saw a very important increase in their cattle populations.Between 1989 and 2000 the number of head of cattle over six months old not destined for fattening increased by 49.8% in the Montaña de Riaño area and by 25.4% in the Montaña de Luna area.This was mainly due to the increase in the number of head for beef production: in the same period the number of beef cattle rose by 5481.8% and 3376.8% in these areas respectively.However, there was also a reduction in the number of cattle farms.Between 1989 and 2000, the number of cattle farms decreased by 48.0% in Montaña de Riaño, while the Montaña de Luna area experienced a 45.1% reduction.This of course implies a notable increase in the mean size of the remaining farms (188.9% and 128.7% respectively) (Junta de Castilla y León, 2002b).
The above changes are not exclusive to these areas; similar processes are underway in other disadvantaged areas of Spain and Europe.Stemming the depopulation and environmental deterioration of these areas requires the development and maintenance of sustainable stock raising systems.This, in turn, requires suff icient information be collected on the status of current stock raising systems plus the changes they undergo, as well as the availability of adequate analytical capacity (Caraveli, 2000;McDonald et al., 2000).
The present work (which forms part of a wider project) reports an attempt to establish a typology that allows the current situation of the cattle farms of the León mountains to be analysed.The process followed can be divided into two stages: the identification of the variables responsible for the differences between farms, and the establishment of homogeneous groups of farms according to these variables (Bisquerra, 1989;Carrasco and Hernán, 1993;Rapey et al., 2001).The main objective of the present work lies within the first of these stages: the identification of the most appropriate variables for characterising and classifying these cattle farms using multivariate and principal components analysis (PCA).

Material and Methods
PCA is a descriptive, factorial, multivariate technique that eliminates redundancy when handling large numbers of variables that are frequently related (Bisquerra, 1989;Carrasco and Hernán, 1993).The process allows the substitution of a large table of quantitative data by one with a smaller number of variables (a linear combination of the originals) known as principal components.The number of principal components obtained can be equal to the number of variables included in the analysis, but from this total a reduced number is selected that explains an acceptable proportion of the overall variance.The number of principal components retained will depend on the phenomenon under study, on the precision required, and on their interpretability (according to the weight of each original variable within the principal component and the correlation between variables and principal components).
The information used in this work was obtained by directly surveying the farms of the Montaña de Riaño area in 1996, 1997 and 1998.The farms studied were all involved in a project, which began in 1996, to produce and market high quality beef from the León mountains.In the first year of the study, 75 farms were surveyed; the number of questionnaires considered adequately completed was 41.In the second and third years, 47 and 45 farms were surveyed; 35 valid sets of information were obtained for each.Thirty three farms were common to every year studied, two were common to the first and second year only, and two were common to the first and third year only.
The information gathered in these surveys, and treated by PCA, refers to production factors (labour, the stock base, land use), the productive characteristics of the farms (types and quantities of produce), and economic aspects (costs, income and profit).
PCA analysis involved the use of the FACTOR and VARIMAX rotation procedures of the SAS statistical package (SAS, 1989).The information collected over the three years of the study was analysed together.Data pertaining to each farm and for each year of the study were taken as single observations, such that PCA was performed with a total of 111 observations (41 for 1996, 35 for 1997 and 35 for 1998).
Table 1 shows the starting set of variables -85 in total.Correlation analysis was performed for these variables to eliminate those providing redundant information.This was undertaken using information gathered in 1998 since this data was considered the most representative of the farms' future prospects.quota-litres milk sold (%): percentage derived from the difference between the number of litres of milk produced by a farm and the number of litres assigned by the quota, over the number of litres assigned by the quota.g % weaned grazing calves, % finished calves, % suckling calves, % replacements, % fattened calves: percentage of calves sold at weaning before fattening and fed only on milk and pasture; for slaughter; not fattened, 1-2 months old; replacements animals and those resold after their acquisition to another farm after a period of fattening.h Cattle costs: sum of costs of feed, sanitary products, fuel, electricity, maintenance of installations, cattle purchase, labour, insurance and others associated exclusively with cattle production.i Total costs: the sum of cattle costs plus those associated with other species (sheep, goats and/or horses).j Income from cattle: sum of the income from the sale of calves, adult cattle not for slaughter, adult cattle for slaughter, milk, subsidies and from the estimated variation in the number of animals held by the farm (capitalization of livestock).k Total income: sum of income from cattle and of other species or the sale of agricultural products.l GM-total: total gross margin-the difference between total income and total costs.m GM-cattle: gross margin for cattle-difference between income from cattle and cattle costs.n Income from calf capitalization: estimated variation in number of calves for one year.o I. capitalization adults: estimated variation in number of breeding cows and studs for one year.Firstly, variables expressing the percentage costs and incomes of the farms were eliminated since these were strongly correlated to others expressed in terms of the number of breeding cows (r ≥ 0.50 and p < 0.005 in all cases).The following were also eliminated: litres milk sold/breeding cow which correlated with income from milk/breeding cow (r = 0.98; p < 0.001); mother cows (%), which correlated with income from subsidies/ breeding cow (r = 0.91; p < 0.001); % suckling calves, which correlated with % finished calves (r = -0.86;p < 0.001); dead calves/calves sold (%), which correlated with dead calves/calves born (%) (r = 0.90; p < 0.001); forage purchased/breeding cow, which correlated with forage costs/breeding cow ( = 0.85; p < 0.001); feed costs/breeding cow which correlated with concentrates/breeding cow (r = 0.99; p < 0.001); feed costs/breeding cow, which correlated with concentrated costs/breeding cow (r = 0.96; p < 0.001); total costs, which correlated with total income (r = 0.91; p < 0.001); GM-total, which correlated with total income (r = 0.97; p < 0.001); GM-cattle without subsidies/breeding cow, which correlated with GMcattle/breeding cow (r = 0.97; p < 0.001) and GM-cattle without subsidies/AWU-cattle, which correlated with GM-cattle/AWU-cattle (r = 0.94; p < 0.001).

Labour
The initial set of variables was thus reduced to 41.Preliminary PCA was performed using this new set of variables to prof ile the structure of the principal components and to eliminate further variables that provided little information (low communality).The variables eliminated were: pasture costs/breeding cow, stud service costs/breeding cow, insurance costs/breeding cow, other costs/breeding cow, dead calves/calves born (%) and income from cattle not for slaughter/ breeding cow.
The final set of variables examined by PCA was 35 (Table 2).

Results
Table 3 shows the principle components selected by PCA, the variance explained by each, the variables with which there is an absolute variable-principal component correlation of > 0.5, and the degree of significance for each.
Seven axes explained 67.1% of the total variance.Figure 1 shows the variables used in PCA in the plane represented by the intersection of principal components (PC) 1 and 2.

Characterization of principal components
PC 1 PC 1 explains 18.9% of the total variance.Bearing in mind the most outstanding relationships between PC 1 and the variables considered, PC 1 can be defined as dealing with the «orientation towards the production of milk, the intensification of the production system used, and the productivity per breeding cow».
In terms of the significance of PC 1, its positive correlation with variables directly related to the and with variables related to productivity per breeding cow (GM-cattle/breeding cow, income from cattle/ breeding cow).Finally, PC 1 shows a strong correlation with the two variables relative to the breeds in the stock base, Pardas (%) and crossbreeds (%) (positive for the former, negative for the latter).In summary, PC 1 has high values in farms with Parda breed cows and an orientation towards milk production, with high production per breeding cow, and with relatively intense production systems.

PC 2
PC 2 explains 12.2% of the total variance.Given its relationship with the variables considered, PC 2 can be described as dealing with «farm size and labour factor productivity».
PC 2 was strongly and positively correlated with variables indicative of the size of the farm (breeding cows/farm, total income) and with those indicative of labour factor productivity (GM-cattle/AWU-cattle, breeding cows/AWU-cattle).In accordance, it was negatively correlated with social security costs/ breeding cow.
It therefore discriminates between farms according to their size (number of breeding cows), the labour requirements of the production system and labour factor productivity.PC 2 becomes important for farms with large numbers of cows, with production systems that allow the handling of many animals per worker, and with high labour factor productivity.
PC 3 PC 3 can be described as dealing with: «aspects related to general farm management».It explains 9.5% of the total variance and is strongly and positively correlated with: cattle costs/breeding cow, costs of replacement animals/breeding cow, fuel costs/ breeding cow, forage costs/breeding cow, maintenance costs/ breeding cow and sanitary product costs/ breeding cow.
PC 3 therefore characterises farms in terms of a number of different costs.The larger or smaller costs faced by farms differentiates them according to aspects of their functioning, e.g., whether a farm produces its own breeding animals and whether it is self-sufficient in forage.PC 3 acquires high values in farms with high costs for fuel, maintenance of installations, machinery, sanitary products, forage and animal replacement, and in farms where there is insufficient forage to cover needs or where replacement animals have to be bought.

PC 4
PC 4 explains 7.7% of the total variance.Taking into account its relationships with the variables studied, it can be def ined as dealing with «complementary activities».
PC 4 is strongly and positively correlated with two variables indicative of farm development in terms of activities complementary to the production of milk and calves, such as the raising of sheep, goats or horses (LUother species (%)) and the re-selling of cattle not for slaughter (costs of re-sold cattle/breeding cow).It is also strongly and positively correlated to % fattened animals.PC 4 discriminates between farms in terms of whether they undertake activities complementary to milk and calf production, e.g., the fattening of weaned grazing calves bought from other farms, the raising of heifers and replacement cows, and the raising of other species.

PC 5
PC 5 explains 6.9% of the total variance and deals with the «technical and economic efficiency of calf production».PC 5 is strongly and positively correlated with two variables indicative of the efficiency of calf production: calves sold/breeding cow and income from calves/breeding cow.It is also strongly and negatively correlated with income from calf capitalization/ breeding cow.PC 5 therefore acquires high values in farms with many calves sold per breeding cow and with high income from this activity.

PC 6
PC 6 explains 6.8% of the total variance.Bearing in mind its relationship with the variables considered, this can be defined as dealing with «extensive management based on grazing».PC 6 is strongly and positively correlated with the stocking rate per hectare of UAA (LU-total/UAA), with income from sales of adult cows for slaughter per breeding cow (income adults slaughtered/breeding cow) and from the percentage of calves sold as weaned grazing calves (% weaned grazing calves).In addition, PC 6 is strongly and negatively related to calves born/breeding cow.
Taking into account its correlation coefficient with LU-total/UAA, PC 6 takes high values in farms with high stocking rates per unit of UAA.This characteristic is associated with reduced availability of UAA and being able to use large areas of common grazing ground.
According to the most outstanding relationships between PC 6 and the variables within this factor, the above characteristics are related to others such as a high breeding cow replacement rate, low reproductive success, and a high percentage of calves sold as weaned grazing calves.

PC 7
PC 7 explains 5.1% of the total variance.Given its relationship with the variables included in the PCA analysis, this factor can be defined as dealing with «calf finishing».PC 7 was strongly and positively associated with the percentage of calves sold for slaughter (% finished calves), and strongly but negatively associated with the percentage of calves sold as weaned grazing calves (% weaned grazing calves).PC7 therefore discriminates between farms in terms of the percentage of calves that complete the production process, increasing their income in line with the relative importance of the number of calves sold for slaughter.

Discussion
The results show that the farms studied can be characterised in terms of two types of variable: those forming what might be termed basic explanatory factors of the variations between farms, and secondary factors (which complement and complete the former).
The basic explanatory factors are represented by PC 1 and 2, and are defined by variables indicative of the orientation of production, the intensification of the production system and the productivity per breeding cow, the productivity of the labour factor, and farm size.
The above characterisation of farms can be qualified in terms of the characteristics that define the rest of the principal components selected (general management, complementary activities, calf production efficiency, and characteristics associated with the grazing system and the importance of calf fattening).
PC 1 and 2 are defined by variables that might be understood as more general (necessary in any attempt to characterise and classify farms) and as defining the production system (orientation, size etc.) (Milan, 1997;Chatellier et al., 2000;Caballero, 2001).The secondary factors are, however, def ined by more specific variables of the study area and with respect to the exact aims of the research etc.
The PCs selected show a strong degree of similarity with those obtained by other authors who have attempted to classify cattle farms in similar areas.For example, they are very similar to those reported by Olaizola et al. (1995), who used PCA to examine 11 variables concerned with the technical and economic characteristics of 50 cattle farms in the Pyrénées.As a result of using PCA to examine 15 variables concerning the same characteristics of 30 cattle farms in the Haute Loire mountains (France), Dobremez et al. (1990) obtained two main axes that explained the variation between these farms.The first referred to specialisation in milk production, the degree of intensif ication and the individual productivity per breeding cow; the second distinguished farms basically in terms of their degree of modernization.
Though it is difficult to establish similarities with other studies -basically because the starting variables are different-the characteristics that def ined the secondary factors in the present work are also reported to be of importance by other authors.For example, Rodríguez and Alfageme (1996) characterised the cattle farms of the Principality of Asturias (northern Spain) in terms of two principal components defined by the relative importance of grazing in the feed of cows and calves, by the conservation of forage, and by the characteristics of the calves at sale (either fattened or at weaned).
Together, the results show the usefulness of PCA in characterising farms.This type of analysis allows one to select from a large number of variables those of greatest importance in explaining the differences between farms.It should be remembered, however, that in the use of this technique, one must evaluate the practicality and potential of any classification derived from the principal components/variables selected.This requires researchers have a certain empirical knowledge of the realities of such farms.

Table 2 .
Variables considered in principal components analysis a AWU: Annual work units.b LU: Livestock units.c UAA: Utilised agricultural area.d Dif.: difference.e GM: gross margin.

Table 3 .
Factors selected from principal components analysis, the variance explained, the significance of each principal component, and correlation coefficients for each principal component and the variables that characterise them