Featured Content

Prediction of live body weight using length and girth measurements for pigs (Jan 17, 2011)
Bookmark and Share

Weight predictions using body measurements have been used in various species of animals.1,2 Backyard farmers in the Philippines used length and girth measurements to estimate weights of their pigs because they could not afford weighing scales.3 A strong correlation between weight and girth measurements has been reported in calves.4 The main method of determining the weight of animals in the absence of weighing scales is to estimate the weight using a number of body characteristics that are readily measured. Typically, weight is regressed on body measurements to determine a weight-prediction equation.

In rural Western Kenya, pigs serve as a major source of family income; farmers mainly keep native breeds that are usually confined by tethering, but may be allowed to roam freely.5 Farmers rely on family labor to manage the pigs, which are fed on locally available feedstuffs. Receiving sufficient money for pigs sold has been a major challenge affecting smallholder pig farming, according to these farmers. Local traders, usually pork butcher men, travel between farms on bicycles looking for pigs to buy. Rural farmers have no system in place to estimate a pig’s weight. Obviously, the most accurate method of measuring a pig’s weight is by weighing it using a scale, but pig farmers in Western Kenya cannot afford such scales. The only option left is guessing the weight of the pig prior to selling. If farmers underestimate the weight of their pigs, they may settle 1for a price below market value and they subsequently lose money. To the author’s knowledge, prediction of the pig’s body weight using girth and length measurements has not been studied in rural Western Kenya or in other similar settings in East Africa. In this study, we determined and validated models that farmers can use to predict live weights of pigs using body measurements.

Materials and methods
This research was approved by the Director of Veterinary Services in Kenya; the Board of Postgraduate Studies, University of Nairobi; and the Animal Care Committee and the Ethics Board of the University of Guelph.

Study locations
Two districts of Western Province, namely Busia and Kakamega, were purposively selected for the study because of the popularity of rural pig keeping in these areas. In each district, two pig-keeping sublocations were identified. A sampling frame of all small-scale pig keepers in each sublocation was established through the help of the local provincial administration. In 1999, Kakamega District had an estimated population of 603,500 and occupied approximately 17% of Western Province. In the same census, Busia District had an estimated population of 370,600 and occupied approximately 14% of Western Province. The village elders guided the researchers in locating the pig farms and played an important role in creating a strong working relationship between the researchers and the pig farmers.

Selection of study farms
In total, 288 farms were included in the study. Farms within each village were randomly selected, proportional to the total number of farms, to include between 65% and 75% of all farms in each village. For each village, each farm was given a number and the numbers were put in a bag. The numbers were then randomly selected from the bag until sufficient farms had been selected. No pig farmer refused to participate during the initial farm visits.

Weighing and measuring pigs
Most of the pigs on the farms visited were weighed, and body measurements were taken at each visit. Only nursing pigs that the farmer wished to sell and pregnant sows were exempt from being weighed or measured. The records of the 42 pigs removed during data cleaning were missing either a weight or an age measurement for unknown reasons.

Other measurements excluded from analyses were those from pregnant sows (n = 130), pigs too heavy for the scale (n = 16), and pigs too difficult to restrain (n = 14).
The farmer was asked to estimate the age and the weight of each pig, which was then restrained by a member of the research team. Small pigs were held in the restrainers’ arms, while pigs too large to hold were restrained using a hog snare. To ensure accuracy, each pig was restrained in a straight posture before any measurement was taken.
A uniquely numbered ear tag was inserted in each pig’s ear. A measuring tape was then used to determine the body length in centimeters from the midpoint between the ears to the point where the tail joined the body . The girth was measured in centimeters around the pig’s body, just behind the forelegs. For pigs weighing < 10 kg, a small spring scale that measured to a maximum of 15 kg was used to weigh the pig. Larger pigs were weighed with a circular spring scale that weighed to a maximum of 100 kg. Each scale was set to zero before a pig was weighed. Both scales were accurate to 0.1 kg. Small pigs were placed in a basketball net with one end tied together; the net was then suspended from the scale. Larger pigs were suspended with two horse girths that were fitted just in front of the hind legs and behind the forelegs. The scale was suspended from a tree branch and the horse girth was attached to the bottom of the scale using a hook . During the follow-up visits, all pigs that were still on the farm were weighed and their length and girth measurements were taken. New pigs were ear-tagged, measured, weighed, and included in the study. Data were recorded on previously prepared weight-recording sheets .

Pig age categories
The pigs were put into three age categories, representing young pigs before the typical market weight was achieved (≤ 5 months old), those in the typical marketing age but younger than the typical breeding age (5.1 months to 9.9 months), and those of breeding age (≥ 10 months). In cases where the pig farmer could not estimate the age of the pig, length and girth measurements were used to place the pig in its appropriate age category.

Data entry and analysis
Data were entered into MS Access (Microsoft corporation, Microsoft Way, Redmond, Washington) and exported to Stata software (Stata Corporation, College Station, Texas) for statistical analysis. When farmers were asked to estimate the age of their pigs, some estimated the number of months they had owned the pig rather than the actual age of the pig. Age was therefore underestimated by approximately 1.5 months, an assumption made because weaned pigs in the study area were typically purchased between 4 and 8 weeks of age. Data were divided into two unique data sets. The first data set (model dataset) was composed of a random sample of 75% of all pigs that were measured once and one randomly selected observation from each pig that was measured more than once. To select these, all pigs that were measured more than once were sorted by pig identification and date when the measurements were taken; each pig measured more than once was thus ordered by the smallest to the largest weight. For pigs measured twice, the first observation of the first pig and the second observation of the second pig were included in the first data set. This systematic process was repeated until all pigs were represented once in the data set. Pigs weighed three times were also ordered by their smallest, middle, and largest weights. The observations were selected in a systematic manner to include the largest, smallest, and middle weights of the first three pigs and so forth until all pigs were represented once in the dataset. This dataset was used to develop the mathematical weight equation.

The second data set (validation dataset) was composed of the remaining 25% of the observations for pigs weighed once and the remaining observation for pigs weighed twice.6 One other observation was included for pigs measured three times. The smallest, middle, and largest weights were selected for the first three pigs, and this pattern was repeated. Each pig was represented only once in this dataset, which was used to validate the weight model developed using the first dataset. The third or remaining observation for the pigs weighed three times was used neither in the model nor in the validation datasets and was not used for this study.

Mixed linear model analyses with a random effect of village (Stata command by sort age category: xtreg weight girth length, re i[village]) were performed by regressing weight on length and girth and gender of the pig for each age category. An additional single model was developed to assess the overall effect of age, length, and girth on the pig’s weight. Fixed effects were retained in the model if they were significantly associated with weight at P < .05. The residuals for the final models for each dataset were examined to determine whether the assumptions of linear regression were met. The predicted weight for each pig in the three validation datasets was determined using the coefficients developed in the model datasets. These were compared to the actual weight measured on each pig, and the differences were used for descriptive statistics. The actual weight was compared to the predicted weight using a paired t test. Finally, the difference between the actual weight and the farmer’s estimate of the pig’s weight was calculated. The distribution of this calculated difference was compared to the distribution of the difference between the actual pig weight and predicted values from the models. The absolute values of the within-pig differences were compared using a paired t test at a 95% level of confidence.

Pig statistics From a total of 628 pigs, 1042 pig observations were made, but complete length, girth, and weight measurements were available for only 840 pig observations. The numbers of pigs examined over time in Busia varied by the farm visit number: 281 observations were made in the initial visit, 226 in the second visit, and 157 in the third visit. During the second visit to Busia district, three farmers chose not to participate in the study, and three other farmers were not interviewed because they no longer had pigs and also were not available to be interviewed. In the third visit, the number of farmers decreased by 29 because they didn’t own pigs during the second visit and they still had not acquired pigs by the time the third visit was made.
A total of 88 farmers acquired new pigs in the course of this study; 78 new pigs were recorded during the second visit and 45 during the third visit. Most of these farmers (57 of 88; 64%) had acquired one pig, and two of 88 (2%) had acquired four pigs. Of the 151 pigs that were lost to follow-up, 115 (76%) had been sold and 20 (13%) had died, while the remaining six (3%) had been stolen.

Age and pig weight
Among the 840 pig observations, 363 pigs (43%) were aged ≤ 5 months, 305 (36%) were aged 5.1 months to 9.9 months, and 172 (21%) were aged ≥ 10 months. In the course of the study period, 449 pigs were weighed once, 146 were weighed twice, and 33 were weighed three times. Pig weight increased with increasing age. On average, these pigs weighed 12 kg (SD = 6.1), 30 kg (SD = 11.4), and 42 kg (SD = 17.0) by age category, respectively. Only 27 pig observations had missing age information because farmers owning them could not estimate the age of these pigs. Fifty-one percent of the observations were on female pigs. The weight of young female pigs up to 5 months of age was 13 kg (SD = 6.4), with males in the same age category weighing 12 kg (SD = 5.8); females aged 5.1months to 9.9 months weighed 30 kg (SD = 12.2), with males in the same age category weighing 29 kg (SD = 10.6). The weight of adult female pigs was 44 kg (SD = 17.9), whereas adult males weighed 35 kg (SD = 12.6). The distribution of body weight in the whole dataset was skewed to the right because there were fewer pigs in the oldest age category.

Pig weight and body measurements
Descriptive statistics for the body measurements in the three age categories are summarized. The relationship between pig weight and girth measurements for the different age categories is illustrated. The 75th percentile of each measurement for the younger pigs overlapped with that of the next older age category. The 75th percentile for length, girth, and weight for the market-age pigs overlapped with the 25th percentile of measurements for the breeding-age pigs.

Regression equations
The model datasets included a total of 509 pig observations: 229 for pigs = 5 months old, 183 for pigs 5.1 months to 9.9 months old, and 97 for pigs = 10 months old. The mean weights for pigs in the three age categories in this dataset were 11 kg (SD = 5.6; 95% CI, 10.6-12.1), 30 kg (SD = 10.9; 95% CI, 28.2-31.3), and 44 kg (SD = 18.6; 95% CI, 40.6-48.1) for pigs ≤ 5 months old, 5.1 months to 9.9 months old, and ≥ 10 months old, respectively.

Length and girth explained 88% to 91% of the total variation in weight for the three pig-age categories. Including village in the model accounted for 15%, 2%, and 26% of the random variation for young, market-age, and breeding-age pigs, respectively. Sex of the pig was not associated with weight in any of the age categories (P > .05). Model diagnostics using the residuals confirmed the assumptions of the models. For the young pigs, ≤ 5 months old, the regression model results indicated that weight increased by 0.18 kg and 0.36 kg as length and girth increased by 1 cm, respectively. For pigs 5.1 months to 9.9 months old, weight increased by 0.39 kg and 0.64 kg as length and girth increased by 1 cm, respectively. Finally, for breeding pigs, aged ≥ 10 months old, weight increased by 0.36 kg and 1.02 kg as length and girth increased by 1 cm, respectively.

Model validation
The second dataset, which comprised 298 observations, was used to validate the model developed above. This included 123 pigs aged = 5 months, 109 pigs aged 5.1 months to 9.9 months, and 66 pigs aged = 10 months. The mean weights for pigs in the three age categories in the dataset were 14 kg (SD = 6.7; 95% CI, 12.4-14.8), 30 kg
(SD = 11.9; 95% CI, 26.7-31.2), and 39 kg (SD = 14.4; 95% CI, 35.7-42.7) for pigs = 5 months old, 5.1 months to 9.9 months old, and ≥ 10 months old, respectively. Examination of the residuals confirmed that the assumptions for the linear regression model were met. Predicted weight increased with increasing length and girth measurements.

Descriptive statistics for the difference between actual body weight and predicted weight are summarized . Similarly, descriptive statistics for the farmer’s estimate of the pig weight minus the actual weight are presented. The weight predicted by the models was a closer approximation of the pig’s actual weight than the farmer’s estimate.

The predicting models underestimated the actual weight of the pigs < 10 months old by 0.08 to 1.1 kg, and overestimated the weight of pigs ≥ 10 months by 0.04 kg. Farmers underestimated the weight of pigs on average by 3.2 kg (SD = 7.9), 2.9 kg (SD = 24.8), and 8.0 kg (SD = 23.4) for young, market-age, and breeding-age pig categories, respectively. The farmer’s estimate of the weight was lower (P < .05) than the actual weight of the pig for the three pig-age categories. There was no difference (P > .05) between observed weight and the weight predicted by the model. The overall absolute difference between the farmer’s estimate and the actual weight (4.18 kg) was larger than the overall difference between the actual pig’s weight and the weight predicted by the model (0.41 kg). This difference (3.77; CI, 1.57-5.96) was statistically significant (P < .05). The single model overestimated the weight of the pig by 0.73 kg (SD = 4.2).

Development of the pig-weight estimation tool
The three weight-prediction equations were presented to farmers during village farmer-training workshops. Observations by the researchers indicated that pig farmers had a difficult time utilizing the equations (data not shown). Because of this difficulty, three weight-estimation tools corresponding to the three pig-age categories were developed. The three age-specific charts and a single chart representing the overall effects of age, length, and girth measurements on pig weight are available at www.aasv.org. Length (distance from the middle of the head between the ears to the point where the tail attaches the body) is presented on the x-axis of each chart, while girth, measured behind the foreleg of the pig, is presented on the y-axis. As an example, and based on the chart representing all age categories, a pig with a girth measuring 80 cm and a length measuring 100 cm will weigh approximately 38 ± 4.2 kg. Demonstrations on how the charts could be used to estimate the live weight of the pigs were performed during the second farmer training sessions.

Source: AASV

Sponsored Links: Best Yield Soybeans, Feed Efficiency, Hog Handling, Ileitis Research, Precision Farming, Swine Reproduction Software,