En jämförelse mellan några multivariata data-analysmetoder
Very often the interesting variables are explained by several underlying variables and in statistical analyses it is common to study the relationship between variables and groups of variables. Because of this multivariate analysis is commonly used in both science and industry. There are two problem...
| Main Author: | |
|---|---|
| Format: | Others |
| Language: | Swedish |
| Published: |
Umeå universitet, Institutionen för matematik och matematisk statistik
1998
|
| Online Access: | http://urn.kb.se/resolve?urn=urn:nbn:se:umu:diva-51358 |
| Summary: | Very often the interesting variables are explained by several underlying variables and in statistical analyses it is common to study the relationship between variables and groups of variables. Because of this multivariate analysis is commonly used in both science and industry. There are two problem with both univariate and multivariate analyses. One is when the variables are correlated. The other is when the number variables of exceeds the number of observations which makes the matrices algebra used in the analysis impossible to execute. Partial Least Square (PLS) is a method that has been developed to handle these problems. The purpose of this master thesis is to compare PLS with other related multivariate and univariate methods. For this reason I have reviewed different methods and described the theoretical similarities between them. I have also used several methods to analyse several different data sets to see how well the methods perform. The conclusion is that PLS is not a universal method that always performs well. It is simply another statistical method that with advantage can be used in some situations. |
|---|