Title Statističke metode u teoriji krivulja rasta
Title (english) Statistical Methods in Growth Curves Theory
Author Adriana Dragić
Mentor Ivan Papić (mentor)
Committee member Nenad Šuvak (predsjednik povjerenstva)
Committee member Ivan Papić (član povjerenstva)
Committee member Danijel Grahovac (član povjerenstva)
Granter Josip Juraj Strossmayer University of Osijek School of Applied Mathematics and Informatics Osijek
Defense date and country 2023-07-10, Croatia
Scientific / art field, discipline and subdiscipline NATURAL SCIENCES Mathematics Probability Theory and Statistics
Abstract U ovom diplomskom radu upoznali smo se sa nekim nelinearnim modelima krivulja rasta kao što su eksponencijalni i monomolekularni modeli, te razni sigmoidalni modeli. Osim što smo se upoznali sa njihovim općim jednadžbama također smo dali i kratki uvid u primjenu tih modela. Svi oni pokazali su se vrlo značajni u raznim područjima znanosti kao što su medicina, šumarstvo, ekonomija, demografija, poljoprivreda, kemija i druge. Neke od modela kao što su eksponencijalni i model opće
potencije potkrijepili smo konkretnim primjerima iz svakodnevnog života. Eksponencijalni modeli uglavnom se koriste u demografiji dok se sigmoidalni modeli često koriste u poljoprivrednim znanostima, šumarstvu, medicini i kemiji. Grafički smo
prikazali svaki obrađeni sigmoidalni model te uvidjeli da je krivulja svakog modela oblika slova S odakle i potječe naziv za takve modele. Na kraju rada dali smo uvid u metodu procjene parametara nelinearnih modela na primjeru iz šumarstva na temelju podataka odnosa starosti i najveće visine norveške smreke. Općenito se procjena parametara nelinearnih modela vrši metodom najmanjih kvadrata, no zbog složenosti nelinearnih modela ručno računanje procjene parametara je vrlo zahtjevno i
često nemoguće. Stoga se za procjenu parametara nelinearnih modela rasta koriste odgovarajuće procedure od kojih je jedna korištena u ovom radu. Preciznije, procjena parametara modela na primjeru podataka o starosti i najvećoj visini norveške
smreke u razdoblju od 1930. do 1974. godine izvršena je pomoći SAS NLIN procedure izvedene iz Marquardtovog iterativnog algoritma. Uvidjeli smo kako su se na razini značajnosti 0.05 parametri kod monomolekularnog, logističkog i Richardsovog
modela rasta pokazali statistički značajni, dok se na toj razini značajnosti parametar β2, koji upravlja brzinom kojom se zavisna varijabla približava svom maksimumu, kod Weibullovog modela nije pokazao statistički značajan. Kod von Bertalanffyevog
modela niti jedan parametar, osim parametra β0, se nije pokazao statistički značajan na razini značajnosti 0.05. Razlog tome je što je za procjenu parametara važan podatak o početnoj visini koja nije dana u podacima pomoću kojih smo vršili procjenu parametara modela. Nakon procjene parametara modela, modeli su zapisani pomoću procijenjenih parametara te je dan i njihov grafički prikaz. Na kraju rada navedeni su mogući problemi prilikom procjene parametara nelinearnih modela ukoliko se koristi SAS NLIN procedura te kako izbjeći iste.
Abstract (english) In this thesis we got acquainted with some of the nonlinear growth curve models such as exponential and monomolecular models and variations of sigmoidal models. Besides familiarizing with their general equations, we gave a short insight in applications of those models. All the models proved to be very significant in different fields of science such as medicine, forestry, economics, demography, agriculture, chemistry and other. We provided real-life examples for some of the models such as exponential and power law models. Exponential models are generally applied in demographics, while sigmoidal models often find their use in agricultural sciences, forestry, medicine and chemistry. We plotted the graph of each sigmoidal model and shown that the curve of each model is S-shaped, hence the name sigmoidal. In the last chapter
we provided an example of the application of one of the methods for estimation of nonlinear growth model parameters based on the age to maximum height of Norway spruce data. Usually, the estimation of nonlinear model parameters is done using the least squares method. However, due to the complexity of nonlinear models their parameters are very difficult and often impossible to estimate analytically. Therefore, we use adequate iterative procedures for that purpose, one of which is covered in this paper. Precisely, model parameter estimation in the example of age to maximum height of Norway spruce ratio for the period 1930-1974 is done by applying the SAS NLIN procedure derived from the Marquardt iterative algorithm. We noted that the parameters of monomolecular, logistic and Richards models are statistically significant with significance level of 0.05. However, using the same significance level, Weibull model parameter β2, which affects the speed of the dependent variable reaching its maximum, did not prove to be statistically significant. None of the von Bertalanffy model parameters proved to be statistically significant on the significance level of 0.05, except for the parameter β0. This is because the data used for estimation does not contain the information about the initial height, which is of
great importance for accurate parameter estimation. By estimating the model parameters, we arrived to model equations and plotted their graphs. To conclude this paper, we listed potential problems that could occur in the estimation using the
SAS NLIN procedure and how to avoid them.
Keywords
eksponencijalna jednadžba
eksponencijalni model rasta
model opće potencije
monomolekularni model rasta
sigmoidalni modeli
logistički model
Gompertzov model
von Bertalanffyev model
Richardsova krivulja
Weibullov model
metoda najmanjih kvadrata
metoda maksimalne vjerodostojnosti
Marquardtova iterativna metoda
SAS NLIN procedura
parcijalne derivacije
početne vrijednosti
statistička značajnost parametara
95% asimptotski pouzdani interval
Keywords (english)
exponential equation
exponential growth model
power law model
monomolecular growth model
sigmoidal models
logistic model
Gompertz model
von Bertalanffy model
Richards curve
Weibull model
least squares method
maximum likelihood estimation
Marquardt iterative method
SAS NLIN procedure
partial derivatives
initial values
statistical significance of parameters
95% asymptotic confidence interval
Language croatian
URN:NBN urn:nbn:hr:126:801146
Study programme Title: Mathematics; specializations in: Financial and Statistical Mathematics, Mathematics and Computer Science, Industrial and Applied Mathematics Course: Financial and Statistical Mathematics Study programme type: university Study level: graduate Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
Type of resource Text
File origin Born digital
Access conditions Open access
Terms of use
Created on 2023-07-11 12:11:50