Approaches to estimate body condition from slaughter records in reindeer

Long-term fluctuations in population densities of reindeer and caribou are common, where pasture is the limiting resource. Pasture quality affects the nutritional status and production of the animals. Therefore, continuous information about changes in the grazing resources is important when making management decisions. �he ob�ecti�e of this study was to in�estigate different possibilities of using routine and additional slaughter records as body condition indicators, and thereby indicators of pasture resources in the summer ranges of reindeer husbandry. Records from 696 reindeer slaughtered in the winter 2002/2003 were included in the study. We de�eloped a model with carcass weight as body condition indicator and two different models combining fatness, conformation, carcass weight, and body size as body condition indicators. �he results showed age and sex dependent differences between the �ariables, and differentiation of animal age and sex impro�ed the precision of models. Ad�usting weight for body size also impro�ed weight as a body condition indicator in adults. Conformation and fatness had good resemblance to weight and body size ad�usted weight and should preferably be included, together with carcass weight and body size measures, when estimating body condition from carcasses. Our analysis showed that using non-in�asi�e slaughter records is a good and non-expensi�e method of estimating body condition in reindeer.

In reindeer husbandry it is desirable to maximize herd producti�ity and economic gain, and reduce the fluctuations in population densities.Knowledge about ongoing changes in pasture quality is therefore essential for optimizing animal density and herd structure and making other appropriate management decisions.To detect changes in pasture quality, continuous monitoring is essential and there is a need to find indicators of pasture quality that are simple and inexpensi�e to measure.
During the snow-free season the reindeer use wide pasture ranges while eating mainly herbs and grasses.Monitoring changes in �egetation o�er such areas would be difficult, time-consuming and expensi�e.�he amount and nutritional quality of a�ailable food during the snow-free season (summer and autumn) strongly affect the calf growth and the energy in�ested in body reser�es (fat and muscles) of all categories of reindeer (Klein, 1964;1968;1991;Reimers, 1983;Reimers et al., 1983;Kumpula et al., 2002).�hus, change in the a�erage body condition (muscle and fat reser�es) of the herd could be a useful indicator of change in the grazing resource during this period.
Se�eral ways to estimate body composition and body condition in reindeer and other large ungulates ha�e pre�iously been suggested.Body condition score (BCS), bioelectrical impendence analysis and reproducti�e status among females are examples of in vivo body composition estimation (Gerhart et al., 1996;Cook et al., 2001;Kumpula, 2001).BCS and bioelectrical impendence require expertise and ad�anced equipment.In addition, data registration could be stressful for the animals.On the other hand, body condition estimation from carcasses after slaughter does not require extra handling of li�e animals.Pre�iously used indicator measures on carcasses include carcass weight, body size, depth of back fat, abdominal fat, bone marrow fat and kidney fat (Dauphiné, 1971;Adamczewski et al., 1987;Helle et al., 1987;Huot, 1988;Chan-Mcleod et al., 1995;Kofinas et al., 2003).Whereas depth of back fat, abdominal fat and kidney fat are measures that require expertise and special equipment, body size measures can be easily recorded ei-ther before or after slaughter and without any ad�anced equipment.Carcass weight together with classifications of conformation and fatness are routine records at reindeer slaughter in Sweden.Carcass weight as a body condition indicator is howe�er biased by the body size of the animal.One solution to this problem would be to ad�ust the weight for differences in body size.Conformation and fatness are used to �alue the carcass for the meat industry.�hese are sub�ecti�e measures and ha�e not been �alidated as body condition indicators; yet they are at least in theory size independent and more direct measures of body reser�es (muscle and fat) than carcass weight.The body condition of adult females is influenced by whether they ha�e had a calf or not during the preceding year (Rönnegård et al., 2002) and should thus be considered.
Here routine records from slaughtered rein-routine records from slaughtered reindeer were used and supplemented with body size records and information about sex, age and reproducti�e status of the animals.�he ob�ecti�e of the study was to in�estigate different possibilities of using routine and additional slaughter records as body condition indicators, with intended use as pasture quality indicators in reindeer husbandry.Specifically Specifically Specifically this addressed the questions: How can routine slaughter records be used indi�idually or combined to estimate body condition?Can additional records impro�e the estimation?

Data collection
Records from 696 carcasses of slaughtered reindeer from 14 herding districts were included in this study (�able 1).�he animals were distributed among fi�e age and sex categories as follows: 103 female cal�es, 313 male cal�es, 44 female yearlings, 77 male yearlings and female adults.�he recorded carcasses were a random sample of animals slaughtered on occasions at three slaughterhouses during the winter 2002/2003.Age and sex of the animals were determined by �isual examination and, when doubtful, age was determined by examination of teeth according to Nieminen (1994).
Weight and EUROP classifications of the carcasses together with information about herding district and slaughter date were obtained from the slaughterhouses.�he EUROP system for classification of carcasses is a standard system of the European Union including two sub�ec-ti�e 15-degree scales for carcass conformation and carcass fat, respecti�ely (Swedish Board of Agriculture, 2002).Prior to the statistical analyses, the conformation and fatness classifications were transformed to quasi-normally distributed �ariables using a threshold model for calculation of the expected �alue of the underlying �ariable in each class.�he normal score (s) for each class (i) was obtained as within each animal category, where φ is the normal density function e�aluated at thresholds i-1 and i, and Ф the cumulati�e normal distribution function e�aluated at the same thresholds.
In addition, three body size measures (back, radius and �aw length) were recorded on each carcass and the reproducti�e status of adult females was �udged.Back length was measured from the front of the second spinious process of the thoractic �ertebrae to the base of the tail (the dorsal side of sancrum).The length of radius (also including the ulna bone) was measured from olecranon tuber to the lower gliding �oint in carpus.�he radius and back lengths were measured on hanging carcasses in the refrigerator room on the day of slaughter or the day after.Jaw (Mandibula) length was measured from the most oral, medial point at the socket of the first incisor to the most posterior point of processus angularis (Angulus mandibulae), as described by Lang�atn (1977).�he �aw and radius lengths were measured on the animals left side.Jaw length was missing for some of the animals (in total 39) due to lost identification of the head.Back length was missing for one animal.�he repro-ducti�e status of adult females, i.e. if they had had a calf during the last summer, was �udged in three classes (calf, no calf or indefinite) by examining the udder for traces of recent milk production (Gerhart et al., 1997).In total 78 fe-  2000).Scores for the obser�ed �ariables in prediction equations for body condition could also be deri�ed (Lawley & Maxwell, 1971).Furthermore SEM, unlike con�entional regression models took uncertainty in the independent �ariables into account, when estimating relationships (Grace, 2003).This uncertainty, i.e. measurement errors, can cause o�erestimation of error terms of the dependent �ariables and bias in parameter estimation of the model if unattended.
In the linear regression analyses, models with herding district and slaughter season and specific category information (sex for cal�es and age for older females) were fitted to carcass weight of cal�es and to the older females to in�estigate effects of discrimination between different groups of animals.�o estimate the impro�ement of the model by ad�usting weight for body size, separate models were analyzed for each animal category.Body size measures were included one at a time, both as simple and cubic �ariables in the models.Variables with P-�alue >0.2 were excluded and �ariables with P<0.05 in �ype III tests were considered significant.�he impro�ements in precision were e�aluated with respect to reduction of residual standard de�iations (SD).�he effects of reproducti�e status of adult females on weight and classification scores were also tested.PCAs were done indi�idually for each animal category.In the first analysis (PCA-I), carcass weight, body size measures and carcass classification scores were included.In the second (PCA-II), the three body size ad�usted carcass weights were added.�he stopping rules in the PCA when extracting principal components were the latent root criterion (i.e.components with eigen�alues of the correlation matrix below one were excluded) and the scree plot criterion (McGarigal et al., 2000).
In the SEM analyses an initial model was constructed for cal�es and older animals, respec-ti�ely (Fig. 1).�he initial models were based on knowledge about reindeer biology and information gained from analysis with linear models and PCA.Latent �ariables were body condition and body size.�he manifest �ariables fatness score, conformation score and carcass weight were assigned to body condition.Carcass weight, back length, radius length and �aw length were assigned to body size.In the calf models, a causal connection from body condition to body size was assumed.�his causal path aimed to capture that cal�es had gained their body size as well as their body reser�es during the preceding snow-free season.In contrast, a causal relationship between body condition and body size did not seem realistic for the older animals since the main part of their body size gain had occurred during earlier years and a co�ariance relationship was used instead.
Measurement models corresponding to the initial models were constructed.�he measurement models were tested and modified based on Wald test, which indicates the loss of model fit (increased χ 2 ) when excluding a particular parameter (Rayko� & Marcoulides, 2000;SAS Institute Inc, 2006).The corresponding structural models for the cal�es were also tested and modified.
A simplified model (SEM-II) with only body condition as latent �ariable was constructed for each animal category.Although we collected three body measures for our analyses, it is likely that only one body measure will be recorded at slaughter for practical reasons.Manifest �ariables attached to body condition were fatness score, conformation score, carcass weight and either one body size measure or carcass weight ad�usted for a body size measure (Fig. 2).�he effect of the latent �ariable body size was captured through a co�ariance between carcass weight and the used body size measure or body size ad�usted carcass weight.�he model was tested and modified for each animal category.
Factor score regression coefficients (Lawley & Maxwell, 1971;SAS Institute Inc, 2006) were retrie�ed for all models.�he regression coefficients could be used to calculate the actual scores of the latent factors for each obser�ation (Lawley & Maxwell, 1971) when estimating body condition.
Maximum likelihood estimation was used in all SEM analyses.Criteria for a good fit of the SEM were a non-significant P-�alue for χ 2 , and comparati�e fit index (CFI) and non-normed fit index (NNFI) both larger than 0.9 (Bentler & Bonett, 1980;Rayko� & Marcoulides, 2000;Pugesek et al., 2003;SAS Institute Inc, 2006).Furthermore, the standardized residuals of the co�ariance matrixes were assured to be symmetrically distributed around zero and that none or only a few residuals were larger than 2.

Results
Carcass weight and the three body measures (back, radius and �aw length) were closely correlated (0.56 -0.71) for the cal�es and male yearlings and less (but not significantly lower) correlated for the female yearlings and female adults (0.34 -0.45).Correlation of weight with conformation and fat classifications was highest for female yearlings (0.62 and 0.63) and female adults (0.54 and 0.56) and slightly lower for female cal�es (0.45 and 0.44) and male cal�es (0.41 and 0.31) although the differences were not significant.In the male yearling group the corresponding correlations were non-significant.Body size measures were uncorrelated with conformation and fatness classifications for the female adults and yearlings, whereas for cal�es, generally small positi�e correlations were found.Howe�er, among male yearlings conformation was negati�ely correlated with back and radius length.A scatter plot matrix of all animals combined is shown in Appendix A.

Carcass weight as body condition indicator
Discriminating between female and male cal�es as well as between female yearlings and adults had significant effects.Residual SD of weight was reduced by 36% when discriminating between female and male cal�es.Discriminating between female yearlings and adults reduced residual SD with 15%.
All body size measures significantly affected weight and explained the �ariation of the data similarly, except for female yearlings where �aw length was not significant (�able 3).Ad-�usting weight for body size, by including the body size measures in the model, reduced the residual SD of weight by up to 33%.
Herding district had a significant effect on carcass weight in all animal categories except female yearlings (�ype III statistics were for female cal�es P=0.011, male cal�es P=0.0017, male yearlings P=0.0028 and female adults P<0.0001).Neither the slaughter period nor  the reproducti�e status of the adult females significantly affected weight in this data set.
Underlying dimensions of the variables �he PCAs resulted in two extracted principal components in all analyses, explaining 73% of the total �ariation in female cal�es, 77% in male cal�es, 72% in female yearlings, 69% male yearlings and 71% in female adults (Fig. 3).In PCA-I, similar patterns in both the first and second components were found for all animal categories except male yearlings (�able 4).In the first component, all loadings were strongly positi�e, suggesting an underlying o�erall size-related linear component in which all �ariables are indicators of different features that are size and mass related.Howe�er, for female yearlings and adults, the body size measures obtained smaller loadings than the weight and classification scores, implying that body size measures explain less of the �ariation in these categories.�his difference was most apparent for the female yearlings, where the largest body size loading was only 57% of the conformation score loading.In contrast, weight and body size measures differed from the classification scores in the first component of the male yearlings, suggesting a more skeleton size related �ariable.�he loadings of the second component of the male yearlings were high for the classification scores and weight and small for the body size measures, distinguishing body resources from body size.The second component of cal�es, female yearlings and female adults showed difference between the loadings for body size measures and those of classification scores, which can be interpreted as a component capturing �ariation in body condition relati�e to body size.PCA-II, where body size ad�usted weight was added, ga�e similar loadings as PCA-I of the classification scores in both the first and second component.For cal�es, the body size ad�usted weights obtained similar loadings as fatness score and conformation score in both the first and second component.For female yearlings and adults the difference between body size measure loadings and classification score loadings were larger than in PCA-I.

Indicators of body condition
�he initial models of SEM-I (Fig. 1) showed good fit to the data and few modifications were done.For female cal�es no model ad�ustments were needed.For male cal�es an extra path between body condition and radius was added (Table 4).In female yearlings and adults no significant co�ariance was found between body condition and body size and the co�ariance was therefore constrained to 0. �he χ 2 , CFI and NNFI of the resulting models are presented in �able 5.All paths were significantly differed from zero at the 5% le�el.Body condition was found to posi-ti�ely affect body size in the SEM-I calf models (Table 4).On the other hand, for male yearlings there was a negati�e co�ariance of -0.3 ± 0.1 (P<0.05) between body condition and body size.Conformation score had the highest loading on body condition of the manifest �ariables.�he models explained 65% of the �ariance in conformation score of the cal�es (�able 6), and 75-99% of the �ariance was explained in the yearling and adult models.Fatness score had the second highest loading although the size of the loading �aried considerably among animal categories (�able 4).In male yearlings only 13% of the �ariance in fatness score was explained, whereas 72% of the �ariance was explained in the female yearling model (�able 6).Carcass weight had the lowest standardized loading on body condition in all animal categories except male cal�es, where radius length had a lower loading (�able 4).Among the manifest �ariables connected to body size, no general grading pattern was found highest although radius length had the highest degree of explained �ariance in all models (�able 6).
Factor score regression coefficients are presented in Appendix B. �he fit of SEM-II models were good according to the fit indices (�able 5), but few paths had significant loadings (Table 7).Male yearling models with back and radius length fitted poorly with one or more not positi�e definite eigen�alues in the co-�ariance-�ariance matrix of the exogenous �ariables.�he female calf model with radius length also fitted poorly with a high χ 2 and corresponding P=0.03, and large standardized residuals (>2).�he fit differed slightly between the models with different body measures.No body measure had good fit for all animal categories independently of using plain body sizes or body size ad�usted weights.�he loadings were similar regardless of kind of body measure.Back length ga�e best fit for male cal�es and female yearlings, �aw length best results for female cal�es and male yearlings, and radius length best for adult females.Rank order of standardized loadings within the models was the same regardless of animal category.Conformation score was the manifest �ariable with largest loading in all models.Body condition score regression coefficients are presented in Appendix C.

Discussion
�he purpose of this study was to explore the opportunities of using routine and additional slaughter records as indicators of body condition reindeer.Such indicators may ser�e as a tool for monitoring changes of reindeer body condition as a consequence of changes in pasture condition.In the long run, estimates of body condition ha�e potential of being an essential part of adapti�e management of the resource base system.
In the study we showed how to combine weight, conformation and fatness classification of carcasses to estimate body condition among slaughtered reindeer.We also showed that precision of carcass weight alone as body condition indicator can be impro�ed by differentiation of animals into specific age and sex classes and by ad�usting weight for body size among yearlings.We found sex and age dependent differences, important to consider when estimating body condition.
�he differences between the obser�ed animal categories can be explained as consequences of age and sex related biological and social factors.Female yearling and adults differed in size and female yearlings were a more homogenous group than female adults.The larger �ariation among adults might be due to that all animals in this group are not fully grown (Skogland, 1983;Gerhart et al., 1997), or that they ha�e experienced different pasture conditions during their different first years.Another reason for inconsistency among the adults may be differences between females with and without calf the last season.Although females with and without calf did not differ in this study, females clearly spend a large amount of body resources on the calf (Gerhart et al., 1997;Rönnegård et al., 2002).On the other hand, the difference between lactating and barren females might be undetectable this late in season.If present, it is possible that the effect could ha�e been detected by more definite information on whether females ha�e had calf or not.
Male yearlings differed markedly from the other categories and also �aried considerably within the group.One reason for the �ariation was probably a consequence of differences in acti�ity during the preceding rut period, when males generally lose body resources.Males are hence no reliable indicators of the status of the snow-free pasture if slaughtered after the rut.
Among cal�es, female cal�es were more similar to the older females than the male cal�es were.Howe�er, the differences between female and male cal�es mainly concerned body size.�he cal�es are all in an intensi�e growth phase during their first snow-free season, and body size growth as well as gain of body resources is both strongly affected by the nutritional status of the animals during this period.Howe�er, their condition is also clearly affected by the condition of their mothers (Lang�atn, 1977;Skogland, 1983;Rönnegård et al., 2002).
We here showed that keeping separate records of female and male cal�es and of yearlings and adults impro�es carcass weight as a body condition indicator.A way to further im-pro�e body condition estimates, besides separating animal categories, is to ad�ust weight for body size to a�oid confounding effects of body size with body condition in yearlings and adults.�o ad�ust calf carcass weight records for size seems not ad�isable, considering that the body size of cal�es also is gained during the snow-free season and therefore closely correlated with body condition (Lang�atn, 1977;Skogland, 1983;Huot, 1988).Here we used three different body size measures to im-pro�e the estimations of body condition.Our results showed no clear ad�antage for any of the three body size measures.Choice of body size measure can therefore be based on other, e.g.practical, reasons.
Although it is possible to use only carcass weight as body condition indicator, the results of this study imply that conformation and fatness are useful complements.The high percentage of the explained �ariance of these �ariables in the SEM analyses, and the similarity in loadings of the size-ad�usted carcass weights and the carcass classification scores for cal�es in the PCA supports this.Conformation and fatness gi�e indications of energy not in�ested into body size increase but in body resources.Conformation appeared to be a better indicator of body condition than fatness as indicated by the higher loadings of conformation in the SEM analyses.�he difference is howe�er small.Both scores are recorded as a routine in scores are recorded as a routine in Sweden and they can easily be used in combination.
Out of the two SEM models, SEM-I was found to be a better model for this data set, �udged from the significant path loadings and good fit indices.Although the fit indices were good, many path loadings were non-significant in SEM-II.On the other hand, SEM-II may be more con�enient for future use whereas fewer complement records are needed and the model structure is the same for all animal categories.�he regression coefficients presented in Appendix B and C can be used to calculate factor scores of body condition and body size (Lawley & Maxwell, 1971).Howe�er, it is important to remember that these factor scores are based on a small data set restricted to one year and that some of the slaughter occasions where quite late in season (January and February).Future impro�ements of factor scores can be made using data from an earlier part of the slaughter season.For estimating change of latent �ariables in long-term data one can also use either latent growth models or multigroup analyses based on the model presented here (Lawrence & Hancock, 1998;Rayko� & Marcoulides, 2000;Pugesek et al., 2003;Arhonditsis et al., 2006).
The use of slaughter records for detecting changes in body condition and pasture quality in�ol�es potential biases that ha�e to be considered.There is a time-lag effect in the condition of the animals, i.e. the nutritional conditions of pre�ious years as well as during the preceding winter may affect the body condition at the time of slaughter (Reimers, 1983;Skogland, 1983;Helle & Ko�ola, 1994;Lundq�ist, 2007).Depending on the time of slaughter, grazing conditions during autumn and early winter may influence the obser�ed body condition.Other �ariables than pasture condition, mainly weather conditions and insect harassment stress, might affect the nutritional status of the animals in the autumn (Reimers, 1983;Helle & Ko�ola, 1994;Lund-q�ist, 2007).Age and sex dependent differences ha�e to be considered since the sensiti�ity of indi�iduals to changes in pasture quality may differ between categories (Adamczewski et al., 1987;Welad�i & Holand, 2003).�he selection of animals to be slaughtered affects slaughter records both directly and indirectly by affecting the li�e reindeer population (Len-�ik, 1988;Rönnegård & Danell, 2003).Slaughter strategies may change o�er years and selection effects can easily cause bias in data in a long-term perspecti�e although, as proposed here, pre�enti�e measures such as registration of age and sex data and a body size indicator can minimize bias.Selection will clearly be less in animal categories where a large part of the animals are slaughtered, as for male cal�es in Sweden.�he calf proportion of total reindeer slaughter has increased during the last 15 years and is at present 63% (Sami Parliament in Sweden, 2007).
The natural resource in reindeer herding is the pasture, and it is important to adapt management actions to changes in the pasture.The management actions ha�e to be adapted to specific changes in each herding district.Slaughter records are a time and cost effecti�e way of data collection and ha�e potential to ser�e as a reliable indicator of changes in reindeer body condition and thereby changes in pasture quality.�his study pro�ided methods to estimate body condition, useful in future management planning.

Fig. 1 .
Fig. 1. �he initial SEM model structure of the first SEM analyses (SEM-I).�he F �ariables are latent �ariables and Vs are manifest �ariables.Ls are the loadings of the paths between the �ariables respecti�ely, Cs are correlation terms, Es are error terms for the manifest �ariables and the D is a disturbance term for the latent �ariable body size.

Fig. 2 .
Fig. 2. Structure of the simplified models (SEM-II).Body condition (F 1 ) is the only latent �ariable, V 1 -V 4 are manifest �ariables.�he Ls are the loadings of the paths between the �ariables, Cs are correlation between �ariables, and Es are error terms for the respecti�e manifest �ariables.

Table 1 .
Distribution of recorded carcasses in herding districts and slaughter houses.2. O�er�iew of data distribution in the fi�e animal categories. �able

Table 3 .
Effect in reduced standard de�iation (SD) by ad�usting weight for body size measurements.

Table 4 .
Comparison of loadings in PCA-I and standardized loadings of SEM-I analyses.
Screeplot of explained �ariance in the PCAs.Diamond represent female cal�es, circle is male cal�es, square is female yearlings, star is male yearlings and triangle represent female adults.

Table 6 .
Degree of explained �ariance of dependent �ariables in SEM-I.

Table 7 .
Standardized loadings and latent co�ariances of best models for each animal category in the SEM-II analyses.