ERP Lecture 7

15 flashcards · Other · by Daniel F.

1. factor analysis

   uses the correlation structure amongst observed variables to model a smaller number of unobserved, latent variables known as factors with a minimum loss of information creating a small, summarized set of different (preferably uncorrelated) factors from a large set of variables (many correlated to each other) Extract maximum common variance from variables, and attribute them a common score creating indexes with variables that measure similar things done before your regression analysis

2. Latent variables

   cannot be measured or observed directly can only be inferred indirectly from other directly observed or measured variables e.g. we can only measure indicators of Socio-Economic Status, like profession, income, education

3. uses of factor analysis

   Reduce clutter: Reduce data complexity to a manageable size while retaining as much of the original information as possible Identify / verify patterns: Reveal (hidden) patterns and relationships in the data, to understand the underlying structure of latent factors

4. types of factor analysis

   Exploratory (EFA) Confirmatory (CFA)

5. Exploratory (EFA)

   no predefined structure to verify no certain number of dimensions in the set of variables goals Searching for structures Understanding the groups of measured variables that relate to factors or theoretical constructs Looking for factors to identify Data reduction, decluttering

6. Confirmatory (CFA)

   verify the factor structure of a set of observed variables Factor structure is known or theorized based on previous theories and research goals: confirm existing ideas, research, or measurement Model Fit: Assess the degree to which the data meet the expected structure Test hypothesis about the structure or the number of dimensions underlying a set of variables e.g., you expect that there are 2 dimensions behind a construct, you collect data and you want to verify the expected structure

7. Graphing factors

   coordinates of a variable: factor loading ideally Large coordinate (factor loading) for one axis, and small coordinate (factor loading) for the other axis

8. Factor extraction and eigenvalue

   How "factorabile" is our data, how many factors should there be eigenvalue: how much of the common variance (communality) of the observed variables a factor explains if > 1 explains more variance than a single observed variable importance of a factor Kaiser's criterion: Keep factors with eigenvalue>1 graph on a scree plot

9. Scree plot

   Graphical method to determine number of factors shows the eigenvalues in a downward curve, ordering them from largest to smallest Retain the factors or components to the left of the inflection point

10. Factor rotation

    Interpret factor loadings to identify factors better minimizes the complexity of factor loadings and makes it easy to interpret the structure rotate the axis so that variables load maximally to only one factor have minimal loading on the other factor turning the axes of the factor space (without changing the data) to make the pattern of relationships between variables and factors clearer

11. Factor rotation techniques

    Orthogonal (varimax) Independent/uncorrelated factors May result in loss of valuable information Oblique (oblimin) Allows factors to be correlated Produce more accurate and reproducible solution depends on whether there is a good theoretical reason to suppose that the factors are related or independent. In social sciences, most often we have correlations among factors

12. Subjectivity of factor analysis

    subjective Choices: Number of factors to extract Rotation method (simple structure, interpretability) Factor interpretation Outcome depends on selected variables Based on correlation matrices

13. Reliability Analysis

    assessment of the degree of consistency between multiple measurements of a variable Cronbachʼs alpha

14. Cronbachʼs alpha

    Coefficient of reliability and measure of internal consistency of tests and measures Good>0.8 (Acceptable >0.7) Values substantially lower indicate an unreliable scale Whether a collection of items consistently measures the same characteristic commonly used with survey data, assessment instruments, which frequently ask multiple questions about the same factor Not a measure of validity cannot determine whether the items measure the correct concept

15. How to Factor Loading Matrix

    a) Number of variables: 6 (rows in matrix) b) Number of factors: 2 (columns in matrix) c) Relationship: Variables 4-6 load highly on factor 1 (values 0.8-0.9), variables 1-3 load highly on factor 2 (values 0.7-0.9)