A Better "Weigh" for Horses: Equine Weight Estimation

 

Through the analysis of data gathered on 224 horses over a two year period, three statistical equine weight estimation models were developed utilizing multiple regression analysis.  Girth, height, length, age, breed, and gender were recorded for each horse in the research sample.  Additionally, a visual estimate of each horse’s weight was made prior to obtaining its actual weight on a digital scale.

Weight estimates for each horse were also calculated using three commercially available girth weight tapes.  These were the Purina, Nutrena, and Sure Measure weight tapes.  Additionally a weight estimate was calculated using the Carroll/Huntington formula1.  This information was obtained to test the comparative accuracy of these estimation methods against the statistical models  developed through this research project. 

Comparisons were made between the equine weight estimates using the girth weight tapes, visual estimates, the Carroll / Huntington formula, and the new regression models developed.

Overall accuracy comparisons were made by comparing the adjusted R squared statistics for each estimation technique/model.  The three estimation models built during the project included girth, height and length as independent variables and actual weight as the dependent variable.

The models constructed through this research project reflect higher adjusted R-squared numbers than any of the other existing equine weight estimation models.

INTRODUCTION:

 

The importance of accurate equine weight estimates is unquestionable.  Veterinarians, equine management facilities, stables, and individual horse owners rely on accurate weight information to determine proper medication dosage, feed and nutrition considerations, racing performance, and transportation requirements2.

The problem is that horse weigh scales are not available in most situations to get an accurate weight on a horse.

            There are currently three primary methods available for estimating equine weight when access to scales is not possible:  1) Visual Estimation, 2) Girth Weight Tapes, and 3) Weight estimation formulae, with the most frequently used and most accurate formula to date being the Carroll/Huntington Formula3.


 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 


It was anticipated that improvements upon the accuracy of the existing methods could be made to develop “A Better ‘Weigh’ For Horses”.  Therefore, it was hypothesized that a more accurate equine weight estimation model could be developed through statistical analysis.

MATERIALS AND METHODS:

 

The most difficult and time-consuming portion of this research project has been gathering data on a sufficient number of horses.  The very problem that generated the interest in this project also created a problem with the data gathering phase—That was finding a facility with an equine scale that could be utilized.

In order to develop a more accurate equine weight estimation model, during phase one of this project data was gathered on 59 Oklahoma State University equine research horses, including girth, length, height, age, breed, gender, an estimated weight by visual estimation, and an actual weight on a scale.

This year, permission was obtained from the Board of the Oklahoma State Fair to measure and weigh horses during the various breed shows at the State Fair.  A formal presentation of the project was given to this group and permission was given to proceed with the data gathering process.  State Fair officials were extremely helpful in arranging stall space for research efforts, passes for entry into the Fair, etc.  At the State Fair data similar to that recorded in 1999 was gathered on 165 additional horses.

For phase one of this project three different commercial horse weight tapes were obtained.  These tapes are used by taking a girth measurement and reading an estimated weight off of the tape.  Each horse in the research sample database was measured and weighed with a regular measuring tape in inches.  Measurements of girth, height, and length were recorded.  Before each horse was weighed on the scale, a visual estimate of the horse’s weight was made.  Each horse’s actual weight in pounds was recorded before returning them to their stalls. 

The girth weight tape results were extrapolated by later laying the girth weight estimation tapes on the ground, taping the ends so that they wouldn’t slide.  The inch measurements were then converted into the weight indication on each girth weight tape and the data was recorded.  These same procedures were also utilized in this year’s data gathering efforts.

Data entry was a substantial task because of all the information gathered on each of 224 horses.  Data was entered into SPSS Statistical Software for analysis and model building.

Initially, the correIation coefficients of each variable to actual weight were analyzed and the results were compared using the Pearson Correlation Coefficient.  Girth was the most highly correlated to actual weight with a coefficient of .9520, followed by height at .8886, length at .8578, gender at            -.4385, breed at .4191, and age at .2632.

Initially it was hypothesized that age, breed, and gender would be more highly correlated with actual weight than they were.  This result caused further investigation into how the variables were coded.  It was determined that binary variables should be used on gender and breed and age to improve the correlation.

Recoding the variables made no significant difference in the correlation of these variables to actual weight.  Also, when age, breed, and gender were utilized in regression equations, they were consistently rejected from the models by the parameters established.  Therefore, efforts were focused on model building with girth, height, and length as the variables in the multiple regression analysis models.

Through Multiple Regression Analysis (MRA), and using the variables girth, height and length, a new multiplicative  model was developed by using the natural logarithm of actual weight as the dependent variable, and the natural logarithms of girth, height, and length as independent variables in a linear regression equation.  When converted, this produced model one with an adjusted R Square of .97559.

Using the same regression technique, a model was also built using volume and height.  The volume variable was created by using girth as the circumference of the base of a cylinder and converting girth into a radius number so volume could be calculated. The length of the horse was used as the height of the cylinder in the volume calculation.

Model two was built in phase one of the project, using Girth2 x Length x Height in a power curve formula.  Each of the three models exceed all existing weight estimation models in accuracy using the sample of 224 horses, as measured by the Adjusted R Square statistic.


RESULTS:

Summary Statistics, Model One

Weight = .003584 x Girth 1.630634 x Height .954088 x Length .399860

 

Multiple R                     .98788

R Square                      .97591

Adjusted R Square        .97559

Standard Error               .06387

 

Step  MultR   Rsq    F (Eqn)       SigF        Var.              BetaIn

1       .9771   .9547   4679.440   .000   In:  LNGIRTH      .9771

2       .9850   .9702   3593.580   .000   In:  LNHEIGHT   .3269

3       .9879   .9759   2971.388   .000   In:  LNLENGTH  .1567

 

Summary Statistics, Model Two

Weight = .018534 x Volume .629388 x Height 1.087000

 

Multiple R                     .98668

R Square                      .97353

Adjusted R Square        .97329

Standard Error               .06680

 

Step  MultR   Rsq    F (Eqn)       SigF        Var.              BetaIn

1       .9766   .9537   4573.049   .000   In:  LNVOL          .9766

2       .9867   .9735   4064.203   .000   In:  LNHEIGHT   .3507

 

 

Summary Statistics, Model Three

Weight = .004209 x (Girth2 x Height x Length) .733855

 

Multiple R                     .98561

R Square                      .97142

Adjusted R Square        .97129

Standard Error               .06926

 

Variable                B             SE B          Beta             T             Sig T

G2HTLGTH    .733855     .008448    .985606       86.864      .0000

(Constant)       .004209    .000601                           6.997      .0000

 

 

As demonstrated in the summary statistics above on the two models constructed this year, and the 1999 model, all are more accurate than any other estimation method or formula currently available as measured by the Adjusted R Square against the sample of 224 horses collected to date.

It should also be noted that the three girth weight tapes were not able to estimate a substantial number of horses in the database because the girths of many horses exceeded the maximum number reflected on the tape.  In calculating the Adjusted R Square for the weight tape estimates, these weight tape samples which could not produce an estimate of weight were excluded.  Thus the true accuracy difference between the models created during this research project and those of the weight tapes is actually larger when it is considered that there were 21 horses for which the weight tapes could not calculate a weight estimate. 

CONCLUSIONS / RECOMMENDATIONS:

 

The research hypothesis that a more accurate equine weight estimation model could be developed through statistical analysis was proven. Not only was the phase one model proven to be more accurate than all other equine weight estimation methods tested with the increased sample size, but the two new models created this year in phase two of the project further improved accuracy levels beyond those of last year.

Through Multiple Regression Analysis (MRA), and using the variables girth, height and length, a new multiplicative model was developed by using the natural logarithm of actual weight as the dependent variable, and the natural logarithms of girth, height, and length as independent variables in a linear regression equation.  When converted, this produced the following model:

 

Weight = .003584 x Girth 1.630634 x Height  .954088 x Length  .399860

 

This model produced an adjusted R Square of .97559, compared to .95349 for the next most accurate model, that of the Australian Equine Veterinarians Carroll / Huntington (The first research model built last year produced an adjusted R Square of .97129 on the new sample, and the volume and height power model resulted in an adjusted R Square of .97329).  Through further data collection and analysis, these models will be refined.

The practical implications of this research are enormous.  An equine weight estimation software program has already been created as a result of this research project, which allows a user to type a girth, height, and length measurement into the appropriate boxes, press <Enter>, and instantly have an accurate weight estimate appear on the screen.  This same concept will eventually be incorporated into an equine weight estimation calculator that can be utilized in the field, producing an immediate weight estimate without the necessity of a personal computer or laptop.

This will simplify the use of the mathematical formulae developed through this research project, and give the average horse owner a more accurate option for estimating the weight of their horses than anything currently available.

At the present time, patents are being sought for the new formulae, and grants for funding necessary equipment (portable equine scale) are being pursued so this research can be continued over the next several years.


Acknowledgments

 

PHASE II ASSISTANCE, FALL 2000:

 

Dr. Bob Curley, Professor of Mathematics, University of Central Oklahoma

 

Oklahoma State Fair Board, with special thanks to D.J. Walker

 

Pat Rush, State Fair Horse Show Secretary

 

Oklahoma City Mounted Police, with special thanks to Danny Booth

 

Glenda Bales, Oklahoma Apaloosa Club President

 

State Fair Horse Show Participants who volunteered their time and allowed their 165 horses to be weighed and measured.

 

My parents, Kris and Joe Hapgood for their untiring support

 

PHASE I ASSISTANCE, FALL 1999:

 

Dr. David Freeman, Professor of Animal Science, State Equine Specialist, Oklahoma State University

 

Dr. MacAllister, Professor, Department of Veterinary Clinical Services, Oklahoma State University

 

Clay Dalluge, Herd Manager, Equine Research Herd, Oklahoma State University

 

The following Oklahoma State University Animal Science Graduate Students:  Britton, Matt, Tammy, Wes and Amber

 

Carl Gedon, Herd Manager, Oklahoma State University

 

Dr. Mark Bianchi, D.V.M.

 

Dr. Jennifer Bianchi, D.V.M.

 

Dr. Bob Curley, Professor of Mathematics, University of Central Oklahoma

 

My parents, Kris and Joe Hapgood for their untiring support

 


References

 

[1]       Carroll C.L. , and Huntington P.J., (1988), Body Condition Scoring and Weight Estimation of Horses, Equine Veterinary Journal 20 (1), 21-45.

           

[2]       Lewis Lon D.,(1982), Feeding and Care of the Horse.  Publisher: Lea & Febiger.

 

[3]       Reavell, D.G., (1999), Measuring and Estimating the Weight of Horses With Tapes, Formulae, and by Visual Assessment, Equine Veterinary Education/AE, December 1999, 188-193.

 

[4]       Marcenac, L.N. and Aublet, H. (1964) Encyclopedie du Cheval, Maloine, Paris.

 

[5]       Ensminger, M.E. (1977) Horses and Horsemanship, 5th ed.

 

[6]       Jones, R.S., Lawrence, T.L.J., Veevers, A., Cleave, N. and Hall, J. (1989) Accuracy of prediction of the liveweight of horses and body measurements.  Vet. Rec. 125, 549-553.