Computers and Electronics in Agriculture 192 (2022) 106572

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag

Cattle reference growth curves based on centile estimation: A

GAMLSS approach
Luiz R. Nakamura a, *, Thiago G. Ramires b, Ana J. Righetto c, Rodrigo R. Pescim d,
Fernanda V. Roquim e, Taciana V. Savian f, Dimitrios M. Stasinopoulos g
a
Universidade Federal de Santa Catarina, Departamento de Informática e Estatística, 274 Roberto Sampaio Gonzaga St, Trindade, 88040-380 Florianópolis, Santa
Catarina, Brazil
b
Universidade Tecnológica Federal do Paraná, Departamento de Matemática. 635 Marcílio Dias St, Jardim Paraíso, 86812-460 Apucarana, Paraná, Brazil
c
Alvaz Agritech. 20 Jose Giroldo St, Guanabara, 86050-268 Londrina, Paraná, Brazil
d
Universidade Estadual de Londrina, Departamento de Estatística. 380 Celso Garcia Cid Rd, Campus Universitário, 86057-970 Londrina, Paraná, Brazil
e
Universidade Federal de Lavras. Programa de Pós-Graduação em Estatística e Experimentação Agropecuária. Aquenta Sol, 37200-900 Lavras, Minas Gerais, Brazil
f
Universidade de São Paulo, Escola Superior de Agricultura “Luiz de Queiroz”. Departamento de Ciências Exatas. 235 Pádua Dias Av, 13418-900 Piracicaba, São
Paulo, Brazil
g
London Metropolitan University, STORM School of Computing. 166-220 Holloway Rd, N7 8DB London, United Kingdom

A R T I C L E I N F O A B S T R A C T

Keywords: In this paper we provide an alternative to create reference growth curves for female Hereford cattle breed based
Animal breeding on the generalized additive models for location, scale and shape (GAMLSS) framework. The proposed method­
Hereford cattle ology avoids some known problems present in the quantile regression, such as crossing quantiles and unsuit­
Multiphasic growth curve
ability in extreme centiles. Since GAMLSS are semi-parametric regression-type models, any statistical distribution
Semi-parametric regression model
Statistical learning
can be considered to explain the behaviour of a given response variable (e.g., body weight) and we are able to
model any and all of the parameters of the response variable distribution, it can easily deal with highly complex
growth curves (e.g., presenting multiple cycles), heteroskedasticity and skewed data issues, commonly present in
this field. We apply the abovementioned methodology, showing that the animal age directly affects median,
variability and skewness characteristics of its growth curve. Finally, this approach may be applied in any other
animal or plant growth, and it can be used as a powerful decision-making tool by producers.

1. Introduction parametric method to fit different quantiles of the distribution rather

than only the mean of the body weight population as in standard
The study and understanding of growth curves play an important nonlinear regression. The main drawback with this methodology is that
role in the management and feeding practices and genetic improvement the estimation process may lead to the quantile curves to cross since that
of any species (Do and Miar, 2020). One of the major approaches to estimates for each quantile are determined separately (further infor­
predict and evaluate animal growth is the study of individual growth mation and details, see Waldmann, 2018; Kneib, 2013). As stated by
curves (Mello et al., 2015), mainly by nonlinear regression models Kneib (2013), some attempts are available in the literature to deal with
(Vázquez et al., 2012). Several authors extensively compared different this problem, however these always require additional efforts and may
nonlinear models in order to fit all sorts of growth curves, such as Tutkun not be appropriate depending on the problem situation. Further, centile
(2019), Wen et al. (2019), Paz et al. (2018), Nogales et al. (2017), among curves estimated at the more extreme part of the data can be very
others. The main argument to use such technique is that the parameters unstable.
of these models usually have direct biological meanings (Darmani Kuhi Hence, in this paper, we provide a semi-parametric regression-type
et al., 2010). model alternative to fit reference curves for Hereford cattle, using the
A different approach, based on quantile regression (Koenker, 2005), generalized additive models for location, scale and shape (GAMLSS)
was considered by Nascimento et al. (2019). The authors used this non- framework (Rigby and Stasinopoulos, 2005). The aim here is not to

* Corresponding author.
E-mail addresses: Luiz.RN@gmail.com, luiz.nakamura@ufsc.br (L.R. Nakamura).

https://doi.org/10.1016/j.compag.2021.106572
Received 9 September 2021; Received in revised form 8 November 2021; Accepted 13 November 2021
Available online 21 November 2021
0168-1699/© 2021 Elsevier B.V. All rights reserved.
L.R. Nakamura et al. Computers and Electronics in Agriculture 192 (2022) 106572

Fig. 1. Growth curves of the Hereford cattle breed.

compare nonlinear mixed models extensively used in the literature to fit low forage storage, depleted by the dry season and by the physiology of
individual growth curves, but to provide a flexible methodology to fit forage plants, contributing to a reduction in total nutrient availability,
reference growth curves so livestock producers can establish any enhanced by the rejection of dry forage by the animal itself.
decision-making needed based on growth standard curves throughout
all range of age, avoiding possible inference estimation process as 2.2. Statistical modelling
crossing quantile issues.
Centile estimation is commonly used for estimating age-related dis­
2. Material and methods tribution of growth curves. Growth centile curves for individuals from a
population are usually achieved by two distinct groups of methods: the
2.1. Data set non-parametric method for quantile regression (Koenker, 2005) and the
parametric method pioneered by Cole (1988) and Cole and Green
In this paper, we modelled 55 Hereford female cattle breed growth (1992). In this paper we are using the second method which was further
curves (Fig. 1) from a farm at Rio Grande do Sul State, Brazil, born in extended by some authors in the literature (e.g., Rigby and Stasino­
1999–2001, using their body weights as the response variable, which poulos, 2004; Rigby and Stasinopoulos, 2006).
were collected every 15 days, from birth to 615 days. The original data In fact, Stasinopoulos et al. (2017) showed that any distribution can
had information up to 675 days, however due to a substantial number of be used to fit centile curves for Y against a single explanatory variable X
missing data, the four final measures were removed. Even though these using the GAMLSS framework (Rigby and Stasinopoulos 2005), which
are not new data and have already been analysed through nonlinear became one of the standard tools for this task, mainly in medical sci­
models by Mendes et al. (2008, 2009) and Nakamura et al. (2013), they ences, as can be seen in several applications such as Rigby and Stasi­
were chosen here due to its highly complexity as displayed in Fig. 1. nopoulos (2004), Rigby and Stasinopoulos (2006), Borghi et al. (2006),
The Hereford cattle occupies a prominent position among the beef Onis et al. (2007), van Buuren et al. (2009), Hossain et al. (2016),
breeds produced in the world and have great potential to enhance, e.g., Rozovsky et al. (2018), Flatley et al. (2019), among others.
feed efficiency and reproductive performance (Reimann et al., 2018), GAMLSS are semi-parametric regression-type models in which in­
since it can adapt to a great number of environments and production volves a distribution for the response variable (e.g., body weight) and
systems. Further, these cattle have a highly fertility level, great weight may involve a non-parametric smoothing term for modelling any or all
gain on pasture and high carcass yield index (Mendes et al., 2008). parameters of the distribution as functions of explanatory variables (e.g.,
Hence, there is a great interest in setting reference growth curves for this age) (Stasinopoulos et al., 2018). These models perform better than
animal breed. quantile regression if the assumed distribution of the response variable
There are clearly two different cycles (two sigmoidal curves) in (e.g., body weight) represents well the empirical data (Kneib, 2013),
Fig. 1, where the variance is not constant throughout the different ages, which may not be a problem, since any distribution can be considered in
increasing over time (heteroskedasticity), and the body weight of the this framework (Rigby et al., 2019). A discussion on the comparison of
female cattle: i) increases from ages 0–200 days and 400–500 days and GAMLSS to the quantile regression is given by Rigby et al. (2013).
ii) is almost constant from ages 300–400 days and 500–615 days. Het­ Mathematically, let Y ∼ D(θ), where D is the distribution of the
eroskedasticity issues are often observed in growth data (Strathe et al., response variable and θ is its parameter vector. For instance, letting D(θ)
2010) and these cycles are commonly observed when seasonal effects, e. represent the Box-Cox Cole and Green (BCCG) distribution with θ =
g., dry and rainy seasons, are also presented. According to Fernandes (μ, σ, ν)⊤ , we reduce to the LMS method of Cole and Green (Cole and
et al. (2012), the absence of rain causes a direct impact on pastures, Green 1992) where the response variable Y > 0 is defined by the
restricting the animal growth (constant) for a period and then increases transformation
again with the change of season up to its maximum weight (plateau).
Further, the authors attest that such behaviour is also associated with

2
L.R. Nakamura et al. Computers and Electronics in Agriculture 192 (2022) 106572

Fig. 2. The fitted parameters from the GAMLSS model against age: (a) body weight median μ; (b) coefficient of variation σ and (c) skewness ν.

⎧ [( )ν ]
⎪ 1 Y in the gamlss package (Stasinopoulos and Rigby, 2007) in R software (R
⎪ 1 ,ν ∕
=0
⎪
⎨ σν
−
μ Core Team 2020), g1 (μ) = log(μ), g2 (σ ) = log(σ ) and g3 (ν) = ν, so they
Z= ( ) (1) guarantee that parameter estimates remain within the appropriate range
⎪
⎪ 1 Y
⎪
⎩ log ,ν = 0 (Stasinopoulos et al., 2017).
σ μ
Since GAMLSS framework allows different regression structures for
where Z is assumed to have a truncated standard normal distribution, each parameter of the response variable distribution, i.e., any and all
μ is the median, σ > 0 is the coefficient of variation and − ∞ < ν < +∞ parameters can be modelled, it can easily handle with the well-known
is a skewness parameter. As stated by Hossain et al. (2016), if ν < 1 or heteroskedasticity issues observed in growth data (Strathe et al., 2010).
ν > 1 we have positive or negative skewness, respectively. In terms of a The amount of smoothing applied in each regression structure, i.e.,
GAMLSS model, we may provide different regression structures for each the estimation of the smoothing parameters is achieved by the penalized
of the parameters as follows: g1 (μ) = s1 (x), g2 (σ) = s2 (x) and g3 (ν) = quasi likelihood method (Lee et al., 2006) implemented in the pb()
s3 (x), where g(⋅) represent appropriate link functions and s(⋅) are function available in the gamlss package (more details are available in
nonparametric smoothing functions, such as the P-splines (Eilers and Rigby and Stasinopoulos, 2014).
Marx, 1996; Eilers et al., 2015). By default, using the BCCGo() function Finally, centile curves for Y against age are achieved by using the

Fig. 3. Fitted conditional distribution of body weight, showing how it changes for different ages.

3
L.R. Nakamura et al. Computers and Electronics in Agriculture 192 (2022) 106572

Fig. 4. Worm plots for the fitted GAMLSS model.

LMS method. From (1) we obtain the 100 α centile yα of Y, given by later.
{ Regarding σ (Fig. 2(b)), we note that the coefficient of variation
{
(standard deviation over the body weight mean) is almost constant,
1
μ(1 + σνzα )ν , ν ∕
=0
yα = (2) since log(σ) barely varies around zero, i.e., the coefficient of variation is
μexp(σzα ), ν = 0 approximately one in all range of age. This result is quite expected
where zα is the 100 α centile of the standard normal distribution (Demuner et al., 2017), since we assume that the variability of newborn
(Cole and Green, 1992). cattle weights should be way smaller than the variability of older cattle
weights. That is, for young cattle, the variability of body weight will be
small, and as they become older, this variability will constantly increase.
3. Results and discussion
The fitted model for ν (Fig. 2(c)) indicates that the distribution of
The fitted models for all parameters (μ, σ and ν) based on the body weight is slightly positively skew (as ̂ ν < 1) in any range of age
GAMLSS framework are displayed in Fig. 2. Plots are used to describe apart from newborn cattle. This positive skewness increases up to age
the relationship between age and each of the parameters since we are 300 days and then becomes constant up to age 615 days. This pattern
fitting additive smoothing terms, and in this sense, the coefficients and may be displayed in Fig. 3, where the fitted conditional distribution of
their corresponding standard errors should not be interpreted, as body weight is displayed along different ages.
emphasized by Ramires et al. (2019). In order to verify the adequacy of the fitted model, we provide in
The fitted model for the parameter μ (Fig. 2(a)) indicates that the Fig. 4 the worm plots (van Buuren and Fredriks, 2001) which are used as
median body weight of newly born Hereford cattle increases until age a residual diagnostic tool. Worm plots are built using the normalized
200 days, decreases at a much slower rate from age 200–400 days, in­ quantile residuals (Dunn and Smyth, 1996). In the plot the cases are split
creases from 400 to 500 days, after which μ is roughly linear until age into 16 age intervals with equal number of cases. Worm plots are
615 days. The fitted smoothing function was able to rightly capture the detrended normal Q-Q residual plots for cases in each of the 16 age
pattern discussed in the Data set Section. intervals and their different shapes can indicate whether the assumed
Since we are modelling the median of the body weight, this result distribution for the response variable is reasonable or not (i.e. if the
slightly differs to the ones found by Mendes et al. (2008, 2009) and assumed model is suitable for the data) for a particular age interval: a
Nakamura et al. (2013) that modelled the body weight average through vertical shift, a slope, a parabola or a S shape, indicate a misfit in the
multiphasic nonlinear regression models. The results found in the three mean, variance, skewness and excess kurtosis of the residuals, respec­
papers tend to underestimate some of the model parameters due to the tively (van Buuren and Fredriks, 2001). The 16 plots are read in rows
positive skewness observed on the body weight that will be discussed from the bottom left plot to the top right plot and correspond to the 16

4
L.R. Nakamura et al. Computers and Electronics in Agriculture 192 (2022) 106572

Fig. 5. Observed body weight, with five fitted centile curves (5, 25, 50, 75, 95) from the fitted GAMLSS model, against age.

the generalized additive models for location, scale and shape (GAMLSS)
Table 1
framework in order to propose an alternative to create standard growth
Comparison of nominal and sample percentages lying at or
curves for Hereford female cattle breed. Despite the complex (cyclic)
below each centile for the fitted model.
observed growth curve used, the GAMLSS model was able to provide a
Nominal centile Sample percentages
good fit to the data set in study as stated through the worm plots. The
5 5.32 provided methodology has been proved to be useful to create standard
25 25.48 Hereford female cattle growth curves. Nonetheless, this methodology
50 49.48
may be applied in any study where the researcher is interested in
75 74.56
95 94.94 creating standard animal or plant growth curves.
Funding sources
This research did not receive any specific grant from funding
age intervals given above the worm plot from lowest to highest age agencies in the public, commercial, or not-for-profit sectors.
(Stasinopoulos et al., 2017). Apparently, the BCCG model provides a
reasonable fit to the data set in study since no particular shape can be
seen in any panel and 95% of the points in each plot lie between the Declaration of Competing Interest
elliptical 95% pointwise interval band curves.
Centile curves (100 α = 5, 25, 50, 75, 95) were then obtained using The authors declare that they have no known competing financial
equation (2) and displayed in Fig. 5. As in Rigby and Stasinopoulos interests or personal relationships that could have appeared to influence
(2004) and Hossain et al. (2016), as a final model diagnostic, we provide the work reported in this paper.
the sample percentages lying at or below each of the fitted curves in
Table 1. As we can note, all fitted centile curves based on the BCCG References
model works quite well since nominal and sample percentages are very
Borghi, E., de Onis, M., Garza, C., Van den Broeck, J., Frongillo, E.A., Grummer-
close. Strawn, L., Van Buuren, S., Pan, H., Molinari, L., Martorell, R., Onyango, A.W.,
Hence, it can be concluded that the BCCG model provides a good fit Martines, J.C., 2006. Construction of the World Health Organization child growth
to these data and has very smooth centile curves that may be used as standards: selection of methods for attained growth curves. Stat. Med. 25 (2),
247–265.
reference growth curves for Hereford cattle breed. For instance, based Cole, T.J., 1988. Fitting smoothed centile curves to reference data (with discussion). J. R.
on these curves, a livestock producer can find out if any female cattle is Stat. Soc. A. 151 (3), 385–418.
underweight in any range of age just comparing its weight with the 5% Cole, T.J., Green, P.J., 1992. Smoothing reference centile curves: the LMS method and
penalized likelihood. Stat. Med. 11 (10), 1305–1319.
centile curve and take an action to solve it, e.g., trying a different diet for
Darmani Kuhi, H., Porter, T., López, S., Kebreab, E., Strathe, A.B., Dumas, A., Dijkstra, J.,
this group. Nonetheless, we shall highlight here that any centile curve, i. France, J., 2010. A review of mathematical functions for the analysis of growth in
e., reference growth curve, can be easily obtained through equation (2). poultry. Poult. Sci. J. 66 (2), 227–240.
Demuner, L.F., Suckeveris, D., Muñoz, J.A., Caetano, V.C., Lima, C.G.d., Faria Filho, D.E.
Finally, note that this methodology may be applied in any study
D., Faria, D.E.D., 2017. Adjustment of growth models in broiler chickens. Pesqui.
where the researcher is interested in creating standard animal or plant Agropecu. Bras. 52 (12), 1241–1252.
growth curves, and also including (if necessary) any extra feature or Do, D.N., Miar, Y., 2020. Evaluation of growth curve models for body weight in
characteristic to model the response variable is possible (for further American mink. Animals 10 (1), 22.
Dunn, P.K., Smyth, G.K., 1996. Randomized quantile residuals. J. Comput. Graph. Stat. 5
details, check Stasinopoulos et al., 2017; Ramires et al., 2021). (3), 236–244.
Eilers, P.H.C., Marx, B.D., 1996. Flexible smoothing with B-splines and penalties (with
4. Conclusions comments and rejoinder). Stat. Sci. 11 (2), 89–121.
Eilers, P.H.C., Marx, B.D., Durbán, M., 2015. Twenty years of P-splines. Sort 39 (2),
149–186.
In this paper we use the Box-Cox Cole and Green distribution under Fernandes, H.J., Tedeschi, L.O., Paulino, M.F., Detmann, E., Paiva, L.M., Valadares
Filho, S.d.C., Silva, A.G.d., Azevêdo, J.A.G., 2012. Evaluation of mathematical

5
L.R. Nakamura et al. Computers and Electronics in Agriculture 192 (2022) 106572

models to describe growth of grazing young bulls. Ver. Bras. Zootec. 41 (2), Reimann, F.A., Boligon, A.A., Campos, G.S., Cardoso, L.L., Junqueira, V.S., Cardoso, F.F.,
367–373. 2018. Genetic parameters and accuracy of traditional and genomic breeding values
Flatley, C., Kumar, S., Greer, R.M., 2019. Reference centiles for the middle cerebral for eye pigmentation, hair coat and breed standard in Hereford and Braford cattle.
artery and umbilical artery pulsatility index and cerebro-placental ratio from a low- Livest. Sci. 213, 44–50.
risk population - a generalised additive model for location, shape and scale Rigby, R.A., Stasinopoulos, D.M., 2014. Automatic smoothing parameter selection in
(GAMLSS) approach. J. Matern. Fetal. Neonatal. Med. 32 (14), 2338–2345. GAMLSS with an application to centile estimation. Stat. Methods. Med. Res. 23 (4),
Hossain, A., Rigby, R., Stasinopoulos, M., Enea, M., 2016. Centile estimation for a 318–332.
proportion response variable. Stat. Med. 35 (6), 895–904. Rigby, R.A., Stasinopoulos, D.M., 2006. Using the Box-Cox t distribution in GAMLSS to
Kneib, T., 2013. Beyond mean regression. Stat. Model. 13 (4), 275–303. model skewness and kurtosis. Stat. Model. 6 (3), 209–229.
Koenker, R., 2005. Quantile Regression. Cambridge University Press, Cambridge. Rigby, R.A., Stasinopoulos, D.M., 2005. Generalized additive models for location, scale
Lee, Y., Nelder, J.A., Pawitan, Y., 2006. Generalized Linear Models with Random Effects: and shape. J. R. Stat. Soc. C. 54 (3), 507–554.
Unified Analysis via H-likelihood. CRC Press, New York. Rigby, R.A., Stasinopoulos, D.M., 2004. Smooth centile curves for skew and kurtotic data
Mello, F., Oliveira, C.A.L., Ribeiro, R.P., Resende, E.K., Povh, J.A., Fornari, D.C., modelled using the Box-Cox power exponential distribution. Stat. Med. 23 (19),
Barreto, R.V., McManus, C., Streit Jr, D., 2015. Growth curve by Gompertz nonlinear 3053–3076.
regression model in female and males in tambaqui (Colossoma macropomum). An. Rigby, R.A., Stasinopoulos, D.M., Heller, G.Z., De Bastiani, F., 2019. Distributions for
Acad. Bras. 87 (4), 2309–2315. Modeling Location, Scale and Shape: Using GAMLSS in R. CRC Press, Boca Raton.
Mendes, P.N., Muniz, J.A., Silva, F.F., Mazzini, A.R.A., 2008. Modelo logístico difásico no Rigby, R.A., Stasinopoulos, D.M., Voudouris, V., 2013. Discussion: a comparison of
estudo do crescimento de fêmeas da raça Hereford. Cienc. Rural. 38 (7), 1984–1990. GAMLSS with quantile regression. Stat. Model. 13 (4), 335–348.
Mendes, P.N., Muniz, J.A., Silva, F.F., Mazzini, A.R.A., Silva, N.A.M., 2009. Análise da Rozovsky, K., Barrowman, N.J., Miller, E., 2018. Centile charts for cranial sutures in
curva de crescimento difásica de fêmeas Hereford por meio da função não linear de children younger than 1 year based on ultrasound measurements. Pediatr. Radiol. 48
Gompertz. Cienc. Anim. Bras. 10 (2), 454–461. (5), 701–707.
Nakamura, L.R., Savian, T.V., Dias, C.T.S., Pinto, L.R.M., Mazzini, A.R.A., 2013. Métodos Stasinopoulos, D.M., Rigby, R.A., 2007. Generalized additive models for location scale
multivariados para agrupamento de bovinos Hereford em função de suas curvas de and shape (GAMLSS) in R. J. Stat. Softw. 23 (7), 1–46.
crescimento. Rev. Bras. Biom. 31 (1), 104–115. Stasinopoulos, M.D., Rigby, R.A., De Bastiani, F., 2018. GAMLSS: a distributional
Nascimento, M., Nascimento, A.C.C., Dekkers, J.C.M., Serão, N.V.L., 2019. Using regression approach. Stat. Model. 18 (3–4), 248–273.
quantile regression methodology to evaluate changes in the shape of growth curves Stasinopoulos, D.M., Rigby, R.A., Heller, G.Z., Voudouris, V., De Bastiani, F., 2017.
in pigs selected for increased feed efficiency based on residual feed intake. Anim. 13 Flexible Regression and Smoothing: Using GAMLSS in R. CRC Press, Boca Raton.
(5), 1009–1019. Strathe, A.B., Danfær, A., Sørensen, H., Kebreab, E., 2010. A multilevel nonlinear mixed-
Nogales, S., Calderón, J., Lupi, T.M., Bressan, M.C., Delgado, J.V., Camacho, M.E., 2017. effects approach to model growth in pigs. J. Anim. Sci. 88 (2), 638–649.
A comparison of the growth performance between cattle reared in conventional Tutkun, M., 2019. Growth curve prediction of holfstein-fresian bulls using different non-
systems and in feral conditions. Livest. Sci. 206, 154–160. linear model functions. Appl. Ecol. Environ. Res. 17 (2), 4409–4416.
Onis, M., Onyango, A.W., Borghi, E., Siyam, A., Nishida, C., Siekmann, J., 2007. van Buuren, S., Fredriks, M., 2001. Worm plot: a simple diagnostic device for modelling
Development of a WHO growth reference for school-aged children and adolescents. growth reference curves. Stat. Med. 20 (8), 1259–1277.
Bull. World. Health. Organ. 85 (9), 660–667. van Buuren, S., Hayes, D.J., Stasinopoulos, D.M., Rigby, R.A., ter Kuile, F.O., Terlouw, D.
Paz, C.C.P., Venturini, G.C., Contini, E., Costa, R.L.D., Lameirinha, L.P., Quirino, C.R., J., 2009. Estimating regional centile curves from mixed data sources and countries.
2018. Nonlinear models of Brazilian sheep in adjustment of growth curves. Czech. J. Stat. Med. 28 (23), 2891–2911.
Anim. Sci. 63 (No. 8), 331–338. Vázquez, J.A., Lorenzo, J.M., Fuciños, P., Franco, D., 2012. Evaluation of non-linear
R Core Team., 2020. R: A language and environment for statistical computing. R equations to model different animal growths with mono and bisigmoid profiles.
Foundation for Statistical Computing, Vienna, Austria https://www.R-project.org/. J. Theor. Biol. 314, 95–105.
Ramires, T.G., Nakamura, L.R., Righetto, A.J., Pescim, R.R., Mazucheli, J., Cordeiro, G. Waldmann, E., 2018. Quantile regression: a short story on how and why. Stat. Model. 18
M., 2019. A new semiparametric Weibull cure rate model: fitting different behaviors (3-4), 203–218.
within GAMLSS. J. Appl. Stat. 46 (15), 2744–2760. Wen, Y., Liu, K.e., Liu, H., Cao, H., Mao, H., Dong, X., Yin, Z., 2019. Comparison of nine
Ramires, T.G., Nakamura, L.R., Righetto, A.J., Pescim, R.R., Mazucheli, J., Rigby, R.A., growth curve models to describe growth of partridges (Alectoris chukar). J. Appl.
Stasinopoulos, D.M., 2021. Validation of stepwise-based procedure in GAMLSS. Anim. Res. 47 (1), 195–200.
J. Data. Sci. 19 (1), 96–110.