MCF avoidance—herd management and associated costs

Table 1 provides an overview of the questions and summary statistics that emanated from the herd management questionnaire (the full data set is provided in S1 Table). Ninety percent of herd owners moved their cattle away from the wildebeest calving grounds to substitute pastures to avoid MCF during the calving season. The owners who did not move their cattle away had small herds consisting of less than seven head of cattle and they cited wanting to keep any milk produced at the boma (a traditional Maasai household unit) as the reason for not moving their herds. The substitute pastures to which cattle were moved were on average 21.3 km away from the boma (Fig. 2), took 2.2 days to reach and the cattle remained there for an average of 88 days (Fig. 3). The distance a herd was moved away from the wildebeest calving grounds to reach a substitute pasture area and the total time it spent away from the boma were moderately associated (p-value < 0.03, r = 0.4) (Fig. 4). Only 10% of herd owners employed non-family members to move and manage their cattle away from the boma, with 90% of households involving only members of the family unit. Within these households an average of 2.3 members (19% of the family unit) were recruited to move the cattle. Of the 90% of herds that moved away to avoid MCF, an average of 82% of each herd was moved, whilst 18% (an average of 15 cattle) were kept at the boma. Of the 18% of cattle that remained at the boma, an average of 74% were lactating. Despite lactating cattle being kept disproportionately at the boma and therefore at greater risk of MCF, 61% of all lactating cattle were moved to substitute pastures, resulting in an average of 71% of the total daily milk being produced by cattle that had been moved away. Of this milk only 10% was returned to the boma. Therefore, on average, 64% of the total daily milk was not available to be used by the 81% of the family unit that remained at the boma. The respondents reported that all of this milk was drunk by the herders or the calves, with none being sold or discarded. This constraint on milk allocation between herders and boma imposes costs on the household. To estimate an upper bound on this cost, we use a study of the Maasai community of Kitengela (Kenya), which found that 52% of the average gross annual household income is derived from milk, of which 27% comes from sales and 25% from consumption [12]. Assume briefly that none of the milk from the moved cows is consumed or sold, the loss of 64% of the household’s milk, which, if the Kitengela population is comparable, amounts to a loss of 33% of the household’s income during the calving season. Assuming that the cattle are away for 88 days (or 24% of the year), this upper-bound represents a loss of about 8% of the household’s annual income. Because the herders and calves do consume the milk the actual loss of income will be less than this. The calculation is useful, however, as it provides a limit above which the economic loss is unlikely to be.

When asked for reasons why the herders had chosen to move their cattle away from the wildebeest calving grounds (and their village pasturelands), 94% said that they had moved away to avoid wildebeest, whilst 6% said they moved their cattle to seek better pasture. The quality of pasture, as perceived by the respondents, at the substitute grazing areas (high 42%, medium 55%, and low 3%) was lower than that of the wildebeest calving grounds (high 52%, medium 48% and low 0%). If MCF were no longer a health risk for their cattle 90% of herd owners say they would change their management practises, with 96% stating that they would no longer move their cattle away from the wildebeest calving grounds.

MCF avoidance—health and condition associated costs and the impact on value

We now present the hedonic price regression results, a comparative summary of the impacts of body condition score (BCS) and heart girth (HG) on price, and a comparison of the changes in cattle condition and market value between the treatment and control herds. Table 2 provides a description of the attribute data retained in the hedonic price regression (the full set of questions and data collected by the market sample is given in S2 Table), Table 3 provides summary statistics and price effects for all these variables, whilst Table 4 contains the final hedonic regression results.

Only cattle characteristics that were immediately verifiable by the buyer had a significant impact on price and were retained in the final regression. Other unverifiable characteristics, including whether or not the individual had been immunised against the locally prevalent disease east coast fever (ECF), the reported daily milk yield or number of previous calves, had no effect. The results show that, with all else constant, the market value of an animal begins to decline at about 86 months (7.2 years) and that the characteristics HG, BCS, Age, Male and Heifer are statistically important determinants of market value. Of these variables, only HG and BCS can be impacted by husbandry conditions and are of interest to the discussion regarding impacts of disease avoidance on value. Table 5 provides the elasticities of market price with respect to BCS, HG, and Age, evaluated at the means of the market data. Both HG and BCS have positive effects on price, but the effect of HG on price is ten times larger than that of BCS, and is more strongly significantly different from zero.

The percentage change in the mean values of HG and BCS and mean price P^* that occurred in time periods 1, 2 and 3 are recorded for each herd in Table 6 (the full set of condition data is shown in S3 Table). Below we present the results including only cattle that made it through to the end of the study.

Time period 1: In this period, during which the treatment herd remained on the pastures of the wildebeest calving grounds whilst the control herd was moved approximately 50 kilometres to substitute grazing areas, the change in the control herd mean BCS (− 5.4%) was significantly different (p = 0.001, t = −3.37, d.f = 140) from that of the treatment herd (+ 0.76%). The mean HG for the control herd remained constant, in comparison to the treatment herd which increased by 3.3% (p < 0.001, t = −3.39, d.f = 180). The combined effect of changes in BCS and HG are reflected in the differences in the change in mean price, P^*, which, in time period 1, increased by 5% for the treatment herd compared to a 0.9% decrease for the control herd (p = 0.0015, t = −3.22, d.f = 215). (These sample means and the t-tests for $\overline{{\widehat{P}}_t^{*}}$ do not account for the variance in the predictions themselves, so while the t-tests test for differences in average predicted prices, they do not account for the sampling variation in these individual predicted prices themselves.)

Time period 2: In this period, during which the control herd was grazed entirely on substitute pastures, the treatment herd’s mean BCS increased by 4.2% whilst the control herd’s mean BCS increased significantly more, by 11.9% (p < 0.0001, t = 4.24, df = 196). In the same interval, the control herd’s mean HG increased by 3.4% whilst the treatment herd’s mean HG increased significantly more, by 6.5% (p = 0.0001, t = −3.98, df = 219). The mean price of cattle for both herds increased in this period, however the price increase experienced by the treatment herd (13.5%) was more than twice that of the control herd (6.8%) (p = 0.0001, t = −3.86, d.f = 221).

Time period 3: In this dry season period, during which the control herd had returned from the distant pastures to graze alongside the treatment herd, the mean BCS of the control and treatment herd decreased by 11.7% and 23.8% respectively (p < 0.0001, t = 4.28, df = 220), whilst the HG decreased by 1.3% and 8.6% respectively (p < 0.0001, t = 10.23, df = 223). The mean price decreased for both herds, however the mean price of a cow in the treatment herd decreased much more, decreasing by 23.8% compared to the 11.7% drop for the control herd (p < 0.0001, t = 9.2, d.f = 221).

Occurrence of disease

During the entire study period 3% of the control herd cattle were reported to have had at least one days sickness, compared to over 90% of the treatment herd. Focussing only on the treatment herd: in time period 1 the mean daily percentage of the herd reported to be sick was 4.5%. By the end of time period 2, however, a respiratory infection had entered the herd and the daily mean percentage of the herd reported sick had increased to 7%, and in time period 3 (until the end of the vaccination trial in early July) the mean rose again to 24% (Fig. 5). The cause of the respiratory infection remains unknown, however analysis indicates that it was not MCF. This clinical data is relevant to the discussion that follows regarding the changes in body condition score and price.

Trial setting

The study took place, with the approval and co-operation of the community, on communally owned land (latitude −3.952239, longitude 36.47537) in a mixed-use buffer zone 25–40 km east of Tarangire National Park in a village called Emboreet in northern Tanzania’s Simanjiro District (Fig. 1). The buffer zone is inhabited by transhumant pastoralists and agro-pastoralist communities and is an important dispersal area for wildlife such as wildebeest (Connochaetes taurinus) and zebra (Equus quagga) that seasonally move out of the national park during the wet season (November to April) to calve. The migration of these herbivores is driven by seasonal water resources and by the high levels of nitrogen and phosphorus in the area’s vegetation that result from underlying volcanic soils and which are important for lactating females [26, 31, 32]. These high quality pastures are referred to in this study as the wildebeest calving grounds.

MCF avoidance—herd management and associated costs

The first part of the economic impact assessment aimed to assess MCF avoidance practices and the associated costs incurred by livestock herd owners. A questionnaire survey targeted 16 Maasai livestock owners, selected because they lived close to the wildebeest calving ground and their herds grazed these pastures when wildebeest were not present (Fig. 1). The survey was carried out four times in 2012. The questionnaire, which was delivered in the Maasai language before being translated to English and was trialled with bilingual participants not involved with the trial, asked the respondents about specific MCF-related management decisions and the impact that these decisions had on their households. The data from this survey are summarized both graphically and in tabular form to provide a basis for a qualitative discussion of the impact of management activities on households, including the distributional impacts of MCF avoidance within the household.

MCF avoidance—health and condition associated costs

The second part of the study aimed to estimate the impact of MCF avoidance on cattle market value by comparing the difference in the market value of animals that were moved to avoid MCF (Fig. 1) against cattle that were not moved. A four-step strategy is followed, involving sample identification, data collection, regression analysis, and market value inference.

1. Sample identification: We use data from three sources:
   1. (a). A trial herd (n = 100) from an ongoing MCF vaccine field trial (hereafter called the treatment herd) which, in contrast with traditional Maasai herding practices, was grazed on the wildebeest calving area during the period February—May. All other husbandry conditions were locally typical.
   2. (b). A control herd, comprised of two privately owned herds (n = 100 & 150) selected on the basis of their size (we wanted the total number of control cattle to be approximately 2.5 times the number of the treatment herd), location (based in the village of Emboreet), and representativeness (locally typical breed and husbandry conditions and, importantly, moved away from the wildebeest calving ground during MCF season).
   3. (c). A market sample (n = 185) which represents a set of cattle individually sold at a primary livestock market in the Simanjiro District. Selection was determined by owner willingness to participate. Animal attribute data, analogous to that recorded for the treatment and control herd, was recorded as described below. In addition, the sale price of each animal was collected. All cattle in (a), (b), and (c) were Tanzanian short-horn.
2. Data collection: At four seasonally relevant time points, a broad set of physical attribute data were collected for all cattle belonging to the treatment and control herds and the market sample. The recorded attribute data included heart girth (cm); wither height (cm); body condition score (1 (thin)—5 (fat)); colour; age (months); the number of pairs of incisors; the vaccination status (with respect to east coast fever (ECF), anthrax / black quarter, and lumpy skin disease); sex; if female, whether a heifer; the outcome of the cow’s last calving; if male, whether castrated. All animal attribute data was collected by the same enumerator and, based on preliminary analysis, only a selection of the variables were used in the final regression. The attribute data collection time points were as follows:
   • Time point 1: Early February when typically the rains have begun and fresh new grass has started to grow. This is the beginning of the period when cattle are moved away from the wildebeest calving ground to substitute grazing pastures and represents the start of the trial period and the base-line for this study;
   • Time point 2: Early March, just after the wildebeest calving season, marks the mid-point of the period in which cattle are kept away from the wildebeest calving ground pastures;
   • Time point 3: Early May, the risk period for MCF transmission has passed and cattle managed in a locally typical manner have just returned to graze on the wildebeest calving ground pastures;
   • Time point 4: Mid-November, represents the end of the long dry season.
   
   Collection of data at these specific points enabled changes in non-fixed attribute data to be quantified across three time periods:
   • Time period I (time point 1 to 2): The control herd were moved away from the wildebeest calving ground to substitute pastures; the end of this period marks the beginning of MCF risk.
   • Time period II (time point 2 to 3): The control herd were grazed on substitute pastures throughout this period, whilst the treatment herd remained on the wildebeest calving ground; this is the high risk period for MCF.
   • Time period III (time point 3 to 4): The control herd have returned to graze alongside the treatment herd on the wildebeest calving ground for the duration of the long dry season; this is the post-risk MCF period.
   
   Disease data were also collected: Control herd owners were asked which of their cattle had suffered from ill-health during the preceding time period (or preceding two months for the first time point). Being an experimental herd for which health data was already being recorded, disease data were compiled for the treatment herd every two or three days.
3. Hedonic price estimation: Based on the attribute and sale price data from the market sample we estimated the parameters of a hedonic price regression [9, 11].
4. Market value inference for control and treatment herds: The attribute data from the treatment and control herds were inserted into to the hedonic regression, which provided an estimate of the market value of each animal in each of the herds based on its physical condition [10]. From these estimates, the average differences in market value across herds and across time periods were calculated. Consequently we were able to infer the herd-level impact of having to move cattle away from the wildebeest calving ground.

The attrition rate of control herd cattle was substantial with only 141 out of 250 cattle making it through to time point 4, with most (65%) being lost to follow up after time point 3. Because we are focusing on changes in cattle condition over time, our results below, and in Table 6 on condition and value changes, utilized only those animals who remained in the herd across all three time periods (those who did not disappear from our sample prior to time point 4). However, attrition could in principle affect the outcome of our analysis if animals died or were sold or given away due (at least in part) to differences in condition, which is likely. To assess this we repeated the means tests with the data on lost cattle included in our calculations until lost. For example, if an animal was sold from the control herd in time period 2, we included the changes in its condition in the first time period. We found that the results were not different qualitatively than if all data for lost animals were excluded, and therefore the inferences do not change.

MCF avoidance—impacts on value

The hedonic price regression relates the market sale price (P_i) to a set of the animal’s attributes X_i for a set of cattle i = 1…N sold at market. Hedonic price theory provides little guidance on the specific functional form of the regression relationship, except that market prices for goods are positive, so consider first the general linear price function (1) where g(⋅) and f(⋅) are transformation functions and ${\mathbf{\text{X}}}_i\mathit{\text{β}}={\sum }_{k=1}^K{X}_{ik}{\beta }_k$ is an index function, linear in K attributes and associated parameters β, including an intercept term. For example, if g(P_i) = P_i and f(X_i; β) = X_i β, equation 1 is a linear regression P_i = X_i β + ɛ_i. Nonlinear Box-Cox formulations of g(⋅) and f(⋅) are possible [33–35], of which linear or log-linear functional forms are the most commonly applied special cases. Likelihood Ratio tests based on preliminary Box-Cox regressions reject the linear form and fail to reject using the log of price as the dependent variable (p = 0.60). We therefore apply a log-linear model of the form (2) where the continuous variables in X may in general be natural logarithms of the original variables (which applies in our specific case), and quadratic interaction terms as described below are included to allow for interaction effects of the explanatory variables. Ordinary Least Squares applied to this regression using the market data (after transforming P to ln(P)) provides consistent parameter estimates $\widehat{\beta }$ assuming a correctly specified model.

A full range of explanatory variables (attribute data) were included in X_i β for the first iteration of the hedonic regression (S2 Table). A final hedonic regression specification was selected based on economic theory and step-wise deletions guided by a combination of F- and T- tests for significance, AIC and BIC values. The heteroskedasticity-robust Huber/White/sandwich covariance estimator was used to calculate standard errors. The most important modelling decisions in this regards are as follows:

First, livestock attributes observable at the market are most likely to impact price (unsubstantiated claims of high milk production or vaccination are likely to be discounted by buyers). This hypothesis was supported from initial regressions. We therefore focused on observable attributes: (i) Heart girth (HG), a summary statistic that correlates strongly with skeletal size, is a statistically important explanatory variable for explaining price variation [36, 37]; (ii) body condition score (BCS), a summary measure for the visual determination of nutritional status of an animal [38]; and (iii) Age, a potentially important characteristic because, even if all physical characteristics are perfectly controlled for, the temporal distribution of costs and benefits from owning an animal vary over its lifetime and benefits or costs accrued in the distant future tend to be discounted relative to those accrued in the near future [39]. Indicator variables for sex (male = 1 if male, zero otherwise) and whether a female is a heifer (heifer = 1 if a female has not calved, zero otherwise) were also included.

Aggregate market prices naturally vary over the course of the year, so we included date-specific indicator variables to account for aggregate market price fluctuations. Logarithmic transformations of the continuous variables and a parsimonious set of linear and quadratic terms minimized the AIC and BIC information criteria.

A description of the retained attribute data is given in Table 2. The final regression used for price estimation is: (3)

A cubic polynomial for age was included to best allow for hypothesized curvature in price due to age, because lower-order polynomials alone did not conform to theory, and because the visual relationships between age and market price indicated a sinusoidal relationship. The three age-associated coefficients are jointly significant (F(3,171) = 4.49, p-value = 0.0047).

This final hedonic regression allows calculation of an estimated market price for any given animal on any given day. To do this we simply enter the cattle specific attribute values recorded for the treatment and control herds into Equation 3.

Our principle objective, however, is to quantify the impact on price that is caused by a change in those specific attributes that could potentially be affected by disease avoidance strategies. Consequently, from the list of attributes that made it through to the final hedonic regression, the attributes of interest to our analysis are BCS and HG.

The percentage change in price with respect to a percentage change in BCS (also known as the elasticity of price with respect to BCS) is (4) where %ΔX = (X_j−X_i)/X_i for a change of any variable from X_i to X_j, and $\frac{\%\Delta P}{\%\Delta \mathit{\text{BCS}}}$ represents the percent difference in P with respect to a percent difference in BCS holding all else constant. The relationship between $\frac{\partial \ln (P)}{\partial \ln (\mathit{\text{BCS}})}$ and $\frac{\%\Delta P}{\%\Delta \mathit{\text{BCS}}}$ is asymptotically exact as changes in BCS approach zero, but is approximate for relatively larger changes in BCS (hence the use of ≈ in Equation 4). The coefficients on the indicator variables such as Male, Heifer and the date variables require careful interpretation as well. An unbiased estimator of the percentage change in the transformed dependent variable P with respect to a change in an indicator variable I from 0 to 1 is $\Delta \text{ln}P∕\Delta I=e^{\widehat{\beta }-\widehat{v\left[\widehat{\beta }\right]}∕2}-1$, where $\widehat{v}[\widehat{\beta }]$ is the estimated variance of $\widehat{\beta }$ , calculated using the Delta Method [40–42].

An estimate of percent change in price due to a percentage change in BCS from one time point, t to the next t + 1 (holding all else constant) is derived by multiplying both sides of Equation 4 by %ΔBCS: (5) where $\frac{\partial \ln (P)}{\partial \ln (\mathit{\text{BCS}})}$ is defined in Equation 4. Analogous elasticities for HG and Age can be derived (as first derivatives) from equation 3.

Equation 5 amounts to a difference equation with respect to ln(HG), ln(BCS), and ln(Age), and is useful for estimating the effect of BCS on price holding all else constant (the same holds for analogous difference equations for HG and Age). Note that Male and Heifer do not vary over time and so drop out, and that the date indicators, which were necessary only to control for market conditions for estimation, are set to zero for the following analysis. Consequently the estimated market price correspond to the base case which is the 1^st market sample time point.

As all three variables vary simultaneously, the difference in estimated ln(P) from one time point to the next, due to simultaneous changes in ln(Age), ln(HG), and ln(BCS), is: (6)

Were we to apply equation 6 to estimate the change in price over time for an animal it would include not only the effects of changes in HG and BCS but also the effects of ageing on price. As animals simultaneously age and change body composition over time the effects of Age should be netted out of the price changes to avoid conflating age and condition effects (HG is strongly correlated with total body weight, which tends to increase more for younger cattle in a given environment [29]). As the age structure of the treatment and control herds were very different (the treatment herd was selected for a vaccination trial, while the control herds were maintained for the benefit of households), netting out the affect of differences in age on price is important for isolating the effects of HG and BCS. To do this we subtracted the age effects from both sides, which provides: (7) where ${\mathrm{\Delta }}_t\ln (P_i^{*})={\mathrm{\Delta }}_t\ln ({\widehat{P}}_i)-\left({\widehat{\beta }}_5{\mathrm{\Delta }}_t\ln {(Age)}_i+{\widehat{\beta }}_6{\mathrm{\Delta }}_t\ln {(Age)}_i^2+{\widehat{\beta }}_7{\mathrm{\Delta }}_t\ln {(Age)}_i^3\right)$ is the estimated percent change in expected price due solely to changes in HG and BCS. The right hand side of equation 7 amounts to the difference equation for ln(P) with respect to changes in ln(HG) and ln(BCS) holding Age constant.

As described above, for any variable X, log-differences are approximately equal to percent changes: Δ_tln(X)_i ≈ %Δ_t X_i (the approximation is asymptotically exact as X_{i, t} − X_{i, t−1} becomes small). Therefore, including the percent changes %Δ_t HG and %Δ_t BCS across time points t = (t = 1→2, 2→3, 3→4) for each animal i in equation 7 along with the estimated coefficients provides an estimated percent change in price ($\%{\mathrm{\Delta }}_t\ln ({\widehat{P}}_i^{*})$) due to changes in cattle condition. Comparison of the sample means ($\overline{\%{\mathrm{\Delta }}_t{\widehat{P}}_{i;h}^{*}}=(1/n_h){\displaystyle \sum _{i=1}^{n_h}{\mathrm{\Delta }}_t}\text{ln}({\widehat{P}}_{i;h}^{*})$) for each herd h = (control, treatment), where n_h is the number of animals in herd h, allows a herd-level comparison of changes in price due to changes in the condition of animals in each of the two herds.

Approval

This study was approved by both the Tanzanian Wildlife Research Institute (TAWIRI) and the Commission for Science and Technology (COSTECH, Tanzania). The human subject research was conducted according to relevant international guidelines and was approved by COSTECH (permit nos.2011-213-ER-2005-141 and 2012-318-ER-2005-141). The animal research, which was non-invasive and was conducted according to international guidelines, was permitted by the Tanzanian Wildlife Research Institute (TAWIRI).