Rätt sida femprocentsspärren
(2022) STAH11 20212Department of Statistics
- Abstract
- This bachelor’s thesis in statistics explores frequentist and bayesian methods for parameter estimation and compares their usefulness when applied to opinion polls of party sympathies in the Swedish political system. Through simulation in R, random samples are generated from a population modelled after the Swedish
electorate at various points during the 2010s and 2020s. The party for which each individual would vote, were it election today, is then noted. Sampling is simulated for three different scenarios with different degrees of variation in the population’s party sympathies over seven months’ time.
These samples are used to estimate party sympathies for the population, both with a frequentist method using the central limit theorem... (More) - This bachelor’s thesis in statistics explores frequentist and bayesian methods for parameter estimation and compares their usefulness when applied to opinion polls of party sympathies in the Swedish political system. Through simulation in R, random samples are generated from a population modelled after the Swedish
electorate at various points during the 2010s and 2020s. The party for which each individual would vote, were it election today, is then noted. Sampling is simulated for three different scenarios with different degrees of variation in the population’s party sympathies over seven months’ time.
These samples are used to estimate party sympathies for the population, both with a frequentist method using the central limit theorem to construct a confidence interval, and with two different bayesian methods using the conjugate prior relationship between the multinomial and Dirichlet distributions to construct credibility intervals. The two bayesian methods differ in regards to the weight that the sample data from a previous month has in estimating the parameters at a later time. Through 1000 runs of each method applied to each scenario, we investigate how often the produced confidence interval (for the frequentist method) or credibility interval (for the bayesian methods) encompasses the true parameter values of party sympathy shares at the last month of measurement.
The bayesian model where the weights of all previous months’ observations accumulate over time is found to be inferior to both other methods, infrequently including the true parameter values in its credibility intervals for all three scenarios. The frequentist model is more comparable to the bayesian model where the weight of a given month’s observations deteriorates over time, but the latter model is ultimately shown to be superior in its accuracy regarding the population party sympathy share. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9071912
- author
- Augustsson, Mårten LU and Haggren, Ida LU
- supervisor
- organization
- alternative title
- om frekventistiska och bayesianska metoder för opinionsmätningar
- course
- STAH11 20212
- year
- 2022
- type
- M2 - Bachelor Degree
- subject
- keywords
- Opinion polls, sampling theory, bayesian statistics, multinomial distribution, dirichlet distribution.
- language
- Swedish
- id
- 9071912
- date added to LUP
- 2022-02-24 11:48:08
- date last changed
- 2022-02-24 11:48:08
@misc{9071912,
abstract = {{This bachelor’s thesis in statistics explores frequentist and bayesian methods for parameter estimation and compares their usefulness when applied to opinion polls of party sympathies in the Swedish political system. Through simulation in R, random samples are generated from a population modelled after the Swedish
electorate at various points during the 2010s and 2020s. The party for which each individual would vote, were it election today, is then noted. Sampling is simulated for three different scenarios with different degrees of variation in the population’s party sympathies over seven months’ time.
These samples are used to estimate party sympathies for the population, both with a frequentist method using the central limit theorem to construct a confidence interval, and with two different bayesian methods using the conjugate prior relationship between the multinomial and Dirichlet distributions to construct credibility intervals. The two bayesian methods differ in regards to the weight that the sample data from a previous month has in estimating the parameters at a later time. Through 1000 runs of each method applied to each scenario, we investigate how often the produced confidence interval (for the frequentist method) or credibility interval (for the bayesian methods) encompasses the true parameter values of party sympathy shares at the last month of measurement.
The bayesian model where the weights of all previous months’ observations accumulate over time is found to be inferior to both other methods, infrequently including the true parameter values in its credibility intervals for all three scenarios. The frequentist model is more comparable to the bayesian model where the weight of a given month’s observations deteriorates over time, but the latter model is ultimately shown to be superior in its accuracy regarding the population party sympathy share.}},
author = {{Augustsson, Mårten and Haggren, Ida}},
language = {{swe}},
note = {{Student Paper}},
title = {{Rätt sida femprocentsspärren}},
year = {{2022}},
}