Modelos de regressão para o valor da produção de erva-mate no Rio Grande do Sul

Abstract

Linear regressionmodelsthatsupposenormalresponsevariableswerewidelyuseduntil more flexibletechniqueswereintroduced.Realsituationsareverycomplex,thusitisdifficult that variablescomplywiththeassumptionsoftheclassiclinearmodel,likethenormalityof the responsevariable.Alternativetechniqueswerethenproposed,like:thegeneralizedlinear models (GLM)andgeneralizedadditivemodelsforlocation,scaleandshape(GAMLSS).GLM main advantageinrelationtolinearmodel(LM)technique,becausetheresponsevariablecan followanydistributionoftheexponentialfamily,besidesofthenormaldistribution.Itisalso possible tomodelotherparametersofthedistributionbasedonthecovariables.Byconsidering the relevanceofthepermanentcultureofyerbamatetotheeconomyRS,thequantityofpublic statistics averableaboutthesubjectandthetechniquesofmodellingstatisticmentioned,theof purpose ofthisstudyistoobtainaregressionmodeltoexplainthevariationoftheproduction valueoftheyerbamateintheRS.ThedatasetwasobtainedfromFoundationofEconomyand Statistics. Thisdatasethastheresponsevariable:productionvalueofyerbamateandquan- titativecovariablesassociatedtoitsproductionandcommercializationin2016.Itisaddedto data setqualitativecovariablesreferringtopolesandmicroregionsstudiedlikedummyvari- ables. Thereafter,descriptivestatisticanalysisisdoneinordertoidentifythebehaviorofall variables.Regressionmodelsareobtainedfromclassicaltechnique(LM)tothemoresophisti- cated (GAMLSS).ItisnotedthattheGAMLSSmodelhadthebestfitaccordingwithmodels selection criteriaandgraphicalanalysisoftheresiduals.FinalGAMLSSmodelatthe5%sig- nificance levelthesignificantquantitativecovariableswere:quantityproduced,harvestedarea and areaforharvesting.Thiscovariablespresentpositivecontributiontoresponsevariableac- cording linearrelationshowedthroughthecorrelationanalysis.PoloPlanaltoMissõeswas the onlysignificantanditpresentedpositiveeffecttoproductionvalue,becauseitissecond largeststateproducer.Thesignificantmicrorregionswere:CaxiasdoSul,Erechim,Frederico Westphalen,Gramado-Canela,Guaporé,Lajeado-Estrela,SantaCruzdoSul,SantaRosaeTrês Passos.ExceptingGramado-Canelamicrorregion,theothersmicrorregionspresentednegative effecttodependentvariable.