Friendship Among Horses- Rank and Kinship Matter
H Sigurjónsdóttir,¹ Machteld C van Dierendonck,² and AG Thórhallsdóttir^3
1Iceland University of Education, Stakkahlid, 105 Reykjavik, Iceland  
Voice 354- 5633858     Fax 3545633833      hrefnas@khi.is
²Tolnegenweg 39 - 41, 3776 PT Stroe, The Netherlands m.dierendonck@planet.nl 
³The Agricultural University at Hvanneyri, Iceland   annagudrun@hvanneyri.is

Introduction

To describe the social structure in a group of horses with some confidence, more than one type of association index is preferable (Hinde, 1976; Kimura, 1998; Whitehead and Dufault, 1999). Although bonding is a well known phenomenon among horses (Houpt, 1982; Waring, 1983; Feh and De Mazieres, 1993; Rees, 1993), few detailed studies have been carried out on feral, semi-wild or domestic groups where all individuals were known and where information on association preferences was gathered from different perspectives. The most detailed studies known to us from this point of view are Wells and Goldschmidt-Rothschild’s study (1979) on the Camargue horses, Kimura’s study (1998) on feral Japanese horses and Clutton-Brock et al’s study (1976) on the Scottish Highland ponies where nearest neighbour measurements and grooming patterns were used to measure friendship. Data on play partners was only gathered in the first study and stallions were not present in the third study. 

When in a herd with a stallion, the mares and their offspring allogroom more with members of the family than with other horses (Wells and Goldschmidt-Rothschild, 1979) while this was not the case where stallions were absent. Then the grooming partners came from the same sex-age groups (Clutton-Brock et al, 1976). Grooming partners stayed close to each other in the Highland ponies group while staying close to each other was not correlated with allogrooming preference in the Camargue herd. The males of all ages played more than the females in the Camargue herd and also in the New Forest ponies herd studied by Tyler (1972). In the New Forest herd, males preferred to play with other males while the young females did not show any sex bias.

Dominance relations were apparent and horses of similar rank tended to stay together in all the three studies. Rank was also correlated with age. That has also been documented for other groups, both feral and domestic (Keiper and Sambraus,1986; Rutberg and Greenberg, 1990; van Dierendonck et al, 1995) and the Przewalski horse (Feh, 1988; Keiper and Receveur, 1992) while other studies on the domestic horse have not shown a significant correlation (Houpt et al, 1978; Houpt and Keiper, 1982) although adults were dominant over juveniles.  Aggressiveness is certainly one of the attributes that influences rank. Another factor is residency since the longer a horse stays in a group, the more time he has to form bonds and gain social experience, which is important for strategy planning. Residency has been shown to influence rank among horses (Clutton-Brock et al, 1976; van Dierendonck et al, 1995). 

The effect of kinship on bonding has mainly been discussed in light of mother-offspring relations. In wild and feral horses and wild asses where stallions form harems, bonds between mothers and offspring are clear (Houpt, 1982; Wells and Goldschmidt-Rothschild, 1979; French, 1998) while no such pattern was apparent amongst the Scottish ponies (Clutton-Brock et al, 1976).

 In this paper we discuss the findings of research on the social structure of three groups of  Icelandic horses with no stallions in light of the factors discussed above, i.e. allogrooming, play, dominance rank, spatial associations (two groups) and kinship (two groups).  It is hoped that the findings can be of some value for both horse behaviourists and others interested in social organisation of mammals. 

Methods

The first group, studied in 1996 at Litla- Drageyri, W-Iceland, was composed of 15 adult horses (5-23 years old); 8 geldings and 7 mares. The composition of the other two groups which were studied more extensively in 1997 and 1999 and which partly consisted of the same horses, is typical for many groups nowadays in Iceland, i.e. adult mares, foals, yearlings, adult geldings and unbroken sub-adults but without stallions.

In 1997 the starting group consisted of 34 animals; 12 pregnant mares, 5 barren mares, each with a suckling yearling, 5 adult geldings, 7 sub-adults and during the study period 8 foals were born (one died). All the horses came from the same farm, Skáney, W- Iceland. In 1999, the majority of the horses, or 23 of 31, also came from Skáney and 18 were the same as in 1997. Eight horses came from two other farms. This time there were 14 pregnant mares, 4 barren adult mares, 4 adult geldings, 5 sub-adult males, 4 sub-adult females and 14 foals were born during the study period. In 1997, four 1-year old un-castrated colts were in the group but only one in 1999. (See van Dierendonck et al , this proceedings) 

In 1996, the horses were rotated regularly between three pastures of different size, 5 in each. This way most of the horses spent some time together. To induce competition each group of 5 was given food pellets and the dominance hierarchy was calculated on the basis of the outcome between the dyads. In total 51 such experiments were carried out. Spatial association measurements were done 203 times (the horses within <1m and1-10m were recorded) and all interactions witnessed in the field (40hours) were recorded. Time-budgets of individuals were also estimated from 200 scanning measurements. The horses were weighed regularly. The horses were observed between the 20^th of June and the 20^th of September. 

In 1997 and 1999, the horses were kept in an 8 ha unmanaged enclosure and were given silage. In 1997, the horses were observed from the 6^th of May till the 11^th of June.   From the 16^th of May when the first foal was born the group was observed almost 24 hours a day. In 1999, the observations started on the 3^rd of May and ended on the 22^nd of June. The group was observed almost 24 hours a day from the birth of the first foal on the 8^th of May. In total 488 hrs in 1997 were used in the analyses and 827 hours from 1999. 

 The ad lib sampling method was used (Lehner, 1996) to collect data on  allogrooming, play and aggressive interactions.  The horses were monitored from an observation hut or outside. Many of the horses were used to people but care was taken not to mix with them. The behavioural classes used were: allogroom , play, threat to bite, bite, chase, attack, ears back, kick, threat to kick, supplant, flee, teeth clapping (see McDonnell and Haviland, 1995).  In 1997 data on spatial associations was gathered. In total 687 maps showing who was the neighbour of whom and how close the horses were to each other, were analysed for the group.  Because the horses did not all spend equal time in the group, calibration matrices were constructed to correct the data for each pair of animals.  

For most of the horses the pedigrees for five generations are known, so it was possible to calculate in detail the genetic relation between all animals both in 1997 and 1999.  The method used was to calculate the inbreeding coefficient, F, between all the dyads (Hartl and Clark, 1997).  The Observer^©, Systat and MatMan^© software were used for data analyses. For the statistics the following software were used: permutation t KR and Mantel T (for matrix comparisons); Adjusted Residuals (for individual preferences); t-tests, Spearman’s rho and Pearson’s correlation coefficients, Landau’s H (with dominance) (Sokal and Rohlf, 1981; Lee and Lee, 1982).

Individual preferences were corrected for the chance the individuals could have met. Hypothesis matrices were all corrected for structural zeros and constructed for Sex and age classes (several different options); Kinship (five generations); Farm (place where the animals were located before joining the herd) and Rank distance.

The rank-order was found by constructing both submission and aggression matrices and adding them together. The submission matrices were based on the frequencies of supplanting, fleeing and teeth clapping. The aggression matrices were based on the frequencies of flattening of the ears, threatening to bite, biting and attacking. The horses were ranked according to the number of dominated individuals. At the end the flipping heuristic method was used to define the place between two adjacent horses in case of equal scores (van Dierendonck et al, 1995).  The improved method of de Vries (1998) was used to calculate Landau´s linearity.

Results

For both 1996 and 1997 there were significant correlations between the matrices showing the frequencies of allogrooming for all dyads and spatial associations (1996: Mantel T= 2,77, p<0.01 (<1m) and T=2,29, p<0.01 (1-10m); 1997:  t =0.185, KR=9349, n=48, p<0.001(within 2 horse lengths)). Also, the sub-adults and geldings tend to play with the same individuals as they allogroom with (1997: t=0.207, KR = 413, n=17, p<0.001; 1999 t =0.4141, KR=400, n=13, p<0.001). Friendship amongst these groups could therefore easily be judged from all these perspectives.  

 In both 1997 and 1999, the horses preferred to allogroom with others of the same sex and similar age groups  (1997: t= 0.3826, KR=7436, n=45, p<0.001; in 1999: t=0.3620, KR =7740, n=45, p<0.001). The adult mares showed a clear preference for other adult mares. When analysed in more detail, the data shows that sex and age are not the only deciding factors because the young mares and the adult geldings did not discriminate between the sexes within their social group. In 1999, there was a clear preference for horses from the same farm (n=45, t=0.169, KR=1968, p<0.001).   

The relationships within the Skáney group were stable over the years (with exceptions) since the horses distributed their allogrooming preferences in a similar way over the years (t=0.492, KR=1485, n=21, p< 0.001).

The adult mares did not play. The 4 year-old geldings played most or on average 5-6 times per 24 hour, both in 1997 and 1999. The sub-adult females played less than the males. The difference between sub-adult males and sub-adult females was significant both in 1997 (t = 4.91, df= 13, p <0.01) and in 1999 (t= 4.86, df= 7, p <0.05). In 1999, the male foals played more than the female foals (t= 2.577, df=11, p<0.05- data too limited in 1997)

The horses preferred to play with others of similar age and sex (1997: t = 0.529, KR=3695, n=24, p<0.001, 1999: t = 0.235, KR=2431, n=17, p<0.001). However, it is clear that this overall correlation does not tell the whole story.  Thus, none of the adult geldings had other adult geldings as significant play-partners; instead their playmates were sub-adult males (only one case of a sub-adult female partner). The young geldings and the yearling stallions preferred to play with each other and to some extent with the adult geldings, but in no cases did they have females as significant play partners. The young mares formed relationships with the males and with each other.  

   The horses tend to be faithful to their playing partners.  Seven horses were present in both years at Skáney and of the 42 potential dyads, 13 were significant in 1997 and 12 in 1999.  Ten of the 1997 partners were also partners in 1999 or 83%.  (See Sigurjónsdóttir et al, submitted.)

The rank orders were significantly linear in all three groups. The rank orders in the two groups at Skáney were positively  related with age, with the older horses dominating the younger (1997: Spearman’s Rho=0.882, n=34, p<0.01; 1999: Rho=0.840, n=31, p<0.01) but not in 1996 when it was on the other hand correlated with weight (r= 0.511, p< 0.05) and how aggressive the horses are as judged by their owners (r=0.588, p<0.05). Interestingly in 1996, the three top ranking horses spent less time grazing than the others and there was a significant negative correlation between weight and time spent grazing (r=-0.753, p< 0.001). 

In 1997 the horses preferred to groom with those who were close in rank (t= 0.268, Kr = 6901, n=41, p<0.001), while in 1999 this correlation was not found. The difference between the years may be due to a strong affiliation between horses from the same farm, masking other effects (in 1999 the horses came from 3 farms). In 1996 the horses who were close in rank tended to be spatially associated (Mantel T = 2.81, p<0.01 for <1m, Mantel T = 3.36, p<0.01 for <1-10m) while that was not the case in 1997 for the Skáney group. The difference here might also be due to unlike group composition. 

Kinship had a clear effect on bonding, both amongst the mares and within the geldings and sub-adult group in 1997 and 1999. (Mares 1997: t= 0.196, KR= 992, n=24, p<0.0005; Mares 1999: t= 0.302, KR= 3131, n=32, p<0.0005; Geldings and sub-adults in 1997: t=0.142, KR= 310, n=17, p<0.005; in 1999: t=0.250, KR= 216, n=12, p<0.0005). This is not due to bonds between mothers and offspring nor between siblings. Instead the horses seem to associate most with individuals of same social group and prefer those who are more related. This argument only seems to apply to familiar horses because no significant correlation was found between kinship values and how much the horses allogroomed for horses who were strangers (the test was carried out for those dyads where both values were positive: Spearman’s rho = 0.138, ns, N= 40). 

Furthermore, the more the mares are related, the closer they are in the dominance hierarchy (in 1997: t=0.219, KR= 1357, n=24, p<0.0005; in 1999: : t=0.300, KR= 3975, n=32, p<0.0005  –mares who gave birth have double identity). However, such a pattern was not found within the other social groups in the herds. 

Discussion

In the groups studied here there was a good agreement between standing close to each other and affiliative relationships. Allogrooming was common, and so was  play by adult geldings and the sub-adults. The horses formed bonds with others of the same social class, but familiarity has a strong effect when few horses are introduced into a large group of familiar individuals. Grooming partners were of similar age among the Highland Ponies (Clutton- Brock et al,1976)  and friends stayed close to each other. The other two feral herds, that have been most intensively studied from this perspective, included a stallion. Thus, the basic unit in the herd of the 27 Camargue horses in France, studied by Wells and van Goldschmidt-Rothschild (1979), was a family made up of a mare, her foal, her yearling and her two year old colt. These individuals allogroomed more within the family than with other horses. The Camargue mares did not make many friendly contacts with other mares although they spent much time with them.  In Japan, in a group of 19 feral horses, the allogrooming bonds seemed to be unstable (Kimura, 1998). Also, those that allogroomed most were on average not the same that were spatially associated as judged from nearest neighbour analyses. Allogrooming was not common.  However, in most studies it seems that association indices based on nearest neighbours measurements or by recording which horses are within a certain distance, are good indicators of who the horses allogroom and play with (Clutton- Brock et al, 1976; Jezierski and Gebler, 1984; Wood-Gush and Galbraith, 1987).

In the groups which had relatively many adult mares the mares formed affiliative bonds almost exclusively with other mares, but not with their offspring. Nevertheless, kinship played a role since those the horses preferred to associate with related individuals. The effect of kinship on bonding with other adults within a group of familiar horses is an interesting finding which has not been reported before. Clutton-Brock et al (1976) also found that mothers and daughters in the Highland ponies herd did not seem to bond, but other genetic relations were not calculated so the conclusions that they could draw were limited. Thus, the existence of grooming bonds between mothers and young offspring (one-four year olds), which has been described for wild horses (Houpt, 1982) and between donkey mothers and offspring after the weaning period (French, 1998) are perhaps not common in a herd dominated by adult mares with no stallion.   

Earlier findings where it has been shown that individuals of similar rank form affiliative relationships and stay close to each other are supported in this study. Thus, van Dierendonck et al (1995) found that in a mixed group of Icelandic horses and Shetland ponies in the Netherlands the horses that tended to stay close to each other were close in rank. The same has been reported for Highland Ponies (Clutton- Brock et al, 1976), Camargue horses (Wells and von Goldschmidt-Rotchchild, 1979), a group of mares of light horses (Ellard and Crowell- Davis, 1989) and Japanese feral horses (Kimura, 1998). This undoubtedly stabilizes the group and reduces aggression. 

In the group (1996) which had almost equal number of geldings and mares and no sub-adults the top ranking horses may have excluded the subordinates from the best grazing spots because they spent less time feeding but were nevertheless heavier. That way, the advantage of staying together is obvious for the most dominant horses. 

The social structure which characterized the Icelandic groups is very likely to be the consequence of the absence of stallions since the herd members have more freedom than if stallions are present (Feist and McCullough, 1976). Perhaps the presence of a stallion has the effect on the social structure that the mares allogroom less, form less stable bonds and have a poorly developed social hierarchy.  Feist and McCullough (1976) suggested this 15 years ago when they compared the social structure of Mountain Wild Horses with that of the New Forest Ponies (Tyler 1972).  

Acknowledgements

We thank the farmers at Skáney for their valuable help and our assistants in the field. Bjarki Eldon calculated the genetic relations. Rannis supported the research and so did Iceland University of Education and Agricultural University at Hvanneyri.

References

Clutton-Brock TH, Greenwood PJ, Powell RP (1976) Ranks and relationships in highland ponies and highland cows. Z Tierpsychol  41: 202-216.

De Vries H (1998) Finding a dominance order most consistent with a linear hierarchy: a new procedure and a review. Anim Behav 55: 827-843.

Ellard ME, Crowell- Davis SL (1989) Evaluating equine dominance in draft mares. Appl Anim Behav Sci 24: 55-75. 

Feh C (1988) Social behaviour and relationships of Przewalski horses in Dutch semi- reserves. Appl Anim Behav Sci 21: 71-87.

Feh C, De Mazieres J (1993) Grooming at a preferred site reduces heart rate in horses. Anim Behav 46: 1191-1194.

Feist C, McCullough D (1976) Behavior patterns and communication in feral horses.  Z Tierpsychol 41: 337-371.

French JM (1998) Mother-offspring relationship in donkeys. Appl AnimBehav Sci 60:  253-258.

Hartl DL, Clark AG (1997) Principles of Population Genetics 3^rd ed.  Sinauer Associates, New York.

Hinde RA (1976)  Interactions, relationships and social structure. Man 11:  1-17.

Houpt KA (1982)  Domestic Animal Behavior for Veterinarians and Animal Scientists. The Iowa State University Press, Ames, Iowa.

Houpt K, Keiper R (1982) The position of the stallion in the dominance hierarchy of feral and domestic ponies. J Anim Sci 54: 945-950.

Houpt KA, Law K, Martinisi V (1978) Dominance hierarchies in horses. Appl Anim Ethol 4: 273-283.

Jezierski T, Gebler E, (1984)  Beobachten zum Sozialverhalten von Polnischen Primitiv Pferden. Z Tierzüchtung und Züchtungsbiologie 101(2): 143-152.

Keiper RR, Receveur H (1992) Social interaction of the Przwalski horse (Equus przwalski) in semi-reserves in The Netherlands. Appl Anim Behav Sci 33: 303-318.

Keiper RR, Sambraus HH (1986) The stability of equine dominance hierarchies and the effect of kinship, proximity and foaling status on hierarchy rank. Appl Anim Behav Sci 16: 121-130.

Kimura R (1998) Mutual grooming and preferred associate relationships in a band of free-ranging horses. Appl AnimBehav Sci 59: 265-276. 

Lee JD, Lee TD (1982)  Statistics and Numerical Methods for Biologists. Wiley, London.

Lehner PN (1996) Handbook of Ethological Methods 2 ed. Cambridge University Press, Cambridge.

McDonnell SM, Haviland JCS (1995) Agonistic ethogram of the equid bachelor band. Appl AnimBehav Sci 43:147-188. 

Rees L (1993) The Horse’s Mind. Stanley Paul Ltd, London.

Rutberg AT, Greenberg SA (1990)  Dominance, aggression frequencies and modes of aggressive competition in feral pony mares. Anim Behav 40: 322-331.

Sigurjónsdóttir H, van Dierendonck M, Thorhallsdóttir AG  (Submitted to Behaviour) Social structure in mixed herds of Icelandic horses. 

Sokal RR, Rohlf  FJ (1981) Biometry. Freeman and Co, London.

Tyler S J (1972) The behaviour and social organization of the New Forest ponies. Anim. Behav Monogr 5 (2): 85-196. 

van Dierendonck MC, De Vries H, Schilder MBH (1995) An analysis of dominance, its behavioural parameters and possible determinants in a herd of Icelandic horses in captivityin the Netherlands.  J Zool (3-4): 362-385.

Waring GH (1983) Horse behavior. Noyes Publications, Park Ridge, New Jersey.

Wells SM, van Goldschmidt-Rothschild B (1979) Social behaviour and relationships in a herd of camargue horses. Z Tierpsychol 49: 363-380.

Whitehead H, Dufault S (1999) Techniques for analyzing vertebrate social structure using identified individuals: review and recommendations. Adv Stud Behav 28: 33-74.

Wood-Gush DGM, Galbraith F (1987) Social relationships in a herd of 11 geldings and two female ponies. Eq Vet J 19(2): 129-132.