izpis_h1_title_alt

Proučitev obravnave manjkajočih vrednosti na primeru raziskave EHIS
BERLIC, NIKA (Author), Žiberna, Aleš (Mentor) More about this mentor... This link opens in a new window

.pdfPDF - Presentation file, Download (1,94 MB)

Abstract
Problem manjkajočih podatkov je v statističnih raziskavah zelo pogost. Obravnave manjkajočih vrednosti se moramo lotiti celostno - po korakih, v nikakršnem primeru pa manjkajočih vrednosti ne smemo ignorirati. Na primeru anketne raziskave EHIS iz leta 2014, ki jo izvaja NIJZ, smo s pomočjo simulacije (izvedene na 1000 ponovitvah) želeli primerjati več različnih metod obravnav manjkajočih vrednosti v okviru različnih mehanizmov ter podati priporočila za njihovo ustrezno obravnavo. Najbolj problematični spremenljivki z vidika obsega manjkajočih vrednosti v raziskavi EHIS (2014) sta numerična in ordinalna spremenljivka dohodek (DOH in DohRaz). Diagnoza manjkajočih vrednosti, ki je bila opravljena s pomočjo logistične regresije, je nakazala na prisotnost vsaj mehanizma MAR. Rezultati simulacij so pokazali, da odstotek manjkajočih vrednosti in mehanizem pomembno vplivata na pristranskost ocen. Izkazalo se je, da sta metodi, ki izločata enote, nepristranski, če vrednosti manjkajo v okviru mehanizma MCAR, vendar neučinkoviti, če je odstotek manjkajočih vrednosti velik. Če mehanizem MCAR ne velja, so se kot najbolj priporočljive metode izkazale multiple imputacije. Za analizo na dejanskih podatkih je bila (na podlagi diagnoze manjkajočih vrednosti in ciljev analize) izbrana metoda multiplih imputacij (MI), ki temelji na pristopu MVNI. Zanimivo pa primerjava rezultatov izbrane metode multiplih imputacij z rezultati analize na razpoložljivih podatkih ni pokazala drastičnih razlik med metodama. Ti rezultati in rezultati analize občutljivosti nakazujejo na močen razlog, ki bi bil lahko v šibkosti mehanizma MAR na spremenljivki dohodek (DOH in DohRaz).

Language:Slovenian
Keywords:diagnoza manjkajočih vrednosti, metode imputacij, analiza občutljivosti, analiza pristranskosti, validacija
Work type:Master's thesis/paper (mb22)
Organization:FE - Faculty of Electrical Engineering
Year:2019
Views:324
Downloads:104
Metadata:XML RDF-CHPDL DC-XML DC-RDF
 
Average score:(0 votes)
Your score:Voting is allowed only to logged in users.
:
Share:AddThis
AddThis uses cookies that require your consent. Edit consent...

Secondary language

Language:English
Title:Addressing Missing Data in Quantitative Social Research - EHIS example
Abstract:
The problem of missing data is relatively common in almost all research. It is important that we treat missing data problem comprehensively, following the steps of guidelines. The purpose of the thesis was to compare different methods for addressing missing data, within the framework of various mechanisms, in order to make recommendations for their proper treatment. Simulations (with 1000 repetitions) were performed on a case of EHIS research (2014), conducted by NIPH, where the most problematic variables are income-related. The diagnosis of missing data, which was carried out through logistic regression, indicated the presence of at least MAR mechanism. The results of the simulations showed that the percentage of missing values and the mechanism significantly influences the bias of the estimates. According to the simulations' results we can conclude that the CCA and PD methods are impartial, insofar as values are missing within the MCAR mechanism, but ineffective if the percentage of missing values is high. If the MCAR mechanism does not apply, multiple imputation have been proven as the most recommended methods. We used joint modelling multiple imputation for the analysis of initial data. Interestingly, the comparison of the results of the selected multiple imputation method with the results of the analysis on the available data did not show drastic differences between these two methods. A possible cause could be the weakness of MAR mechanism on the observed variable.

Keywords:missing values diagnosis, imputation methods, sensitivity analysis, bias analysis, validation

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Comments

Leave comment

You have to log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back