unit 10 slides
DESCRIPTION
TRANSCRIPT
Psychology 3800, Lab 002
Factor Analysis
• assignment #8: feedback
• factor analysis: overview
• example analysis in SPSS
• final assignment
In Lab Today…
Assignment 8: Feedback Check the lab blog for a comments and suggestions:
http://uwo3800g.tumblr.com/post/81517223719/assignment-8-commonly-made-errors
Help posts also available on:
(1) Reading the Excluded Variables table: http://uwo3800g.tumblr.com/post/81332757555/faq-how-do-we-interpet-the-information-in-the-excluded
(2) Differentiating between r and R: http://uwo3800g.tumblr.com/post/81575779291/faq-what-is-the-difference-between-r-and-r-in
Factor Analysis: Overview
• method of data reduction (i.e. reduces collection of variables to a smaller number of variable clusters)
• determines the overarching structure of the data by identifying underlying factors within our data
• variables that are highly intercorrelated will be sorted into the same factor
What Is Factor Analysis?
Note: factors are also called “dimensions”, “components” or “latent variables”
• interested in variables contributing to exam grades in Psych 3800
• measured 135 students on 9 variables • attitude toward math • class attendance • office hours visits • previous stats experience • time management skills • level of partying during the week before the exam • tendency to procrastinate • level of studying • effort put into lab assignments
• want to identify whether this list can be reduced to a few broad factors
Example Analysis
Run a factor analysis in SPSS…
• assesses which variables are highly intercorrelated (and uncorrelated with other variables)
• groups variables into minimum number of factors (parsimony) that account for a significant (large) portion of the data
• outputs factor loadings (correlations of each variable with the factors)
• researcher assesses factor loadings to decide on what the factors represent
General Analysis Approach
Example Output
Variables Factor 1 Factor 2 Factor 3
Assignment effort -.157 .629 .388 Class attendance .258 .723 .124 Office hours visits .063 -.663 .206 Math love -.122 -.091 .864 Previous experience .372 .370 .669 Partying -.742 .038 -.272 Procrastination -.753 .296 -.187 Studying .720 .017 -.273 Time management .596 .245 -.130
Example Output
Variables Factor 1 Factor 2 Factor 3
Assignment effort -.157 .629 .388 Class attendance .258 .723 .124 Office hours visits .063 -.663 .206 Math love -.122 -.091 .864 Previous experience .372 .370 .669 Partying -.742 .038 -.272 Procrastination -.753 .296 -.187 Studying .720 .017 -.273 Time management .596 .245 -.130
Identifying which variables load most highly on which factor…
Variables Exam Prep Coursework Experience
Assignment effort -.157 .629 .388 Class attendance .258 .723 .124 Office hours visits .063 -.663 .206 Math love -.122 -.091 .864 Previous experience .372 .370 .669 Partying -.742 .038 -.272 Procrastination -.753 .296 -.187 Studying .720 .017 -.273 Time management .596 .245 -.130
Example Output
Naming the factors to reflect the pattern of factor loadings…
So, the various variables assessed can be grouped into three overarching factors:
1) level of preparation for the exam 2) diligence with various aspects of the course 3) general experience with math/stats
We can now use these latent variables to predict course success rather than assessing one variable at a time
• each factor represents a broader underlying concept • each concept takes more than one variable into account
Example Conclusion
• eigenvalues provide insight into the magnitude of each factor extracted in factor analysis (greater value = greater magnitude)
• so, there will be an eigenvalue for each factor in the analysis
• to calculate the eigenvalue for each factor: o square each factor loading o sum up the factor loadings within each factor (columns)
• note: if we divide each eigenvalue by the number of variables in the factor analysis, we obtain a value representing the proportion of variance in the data accounted for by each factor
Proportion of Variance For Each Factor: Eigenvalues
Variables Factor 1 Factor 2 Factor 3
Assignment effort -.1572 .6292 .3882 Class attendance .2582 .7232 .1242 Office hours visits .0632 -.6632 .2062 Math love -.1222 -.0912 .8642 Previous experience .3722 .3702 .6692 Partying -.7422 .0382 -.2722 Procrastination -.7532 .2962 -.1872 Studying .7202 .0172 -.2732 Time management .5962 .2452 -.1302
Proportion of Variance For Each Factor: Eigenvalues
squaring each factor loading from original output…
Variables Factor 1 Factor 2 Factor 3
Assignment effort -.1572 .6292 .3882 Class attendance .2582 .7232 .1242 Office hours visits .0632 -.6632 .2062 Math love -.1222 -.0912 .8642 Previous experience .3722 .3702 .6692 Partying -.7422 .0382 -.2722 Procrastination -.7532 .2962 -.1872 Studying .7202 .0172 -.2732 Time management .5962 .2452 -.1302
Eigenvalue Σ = 2.240 Σ = 1.653 Σ = 1.603
Proportion of Variance For Each Factor: Eigenvalues
magnitude of each factor
Variables Factor 1 Factor 2 Factor 3
Assignment effort -.1572 .6292 .3882 Class attendance .2582 .7232 .1242 Office hours visits .0632 -.6632 .2062 Math love -.1222 -.0912 .8642 Previous experience .3722 .3702 .6692 Partying -.7422 .0382 -.2722 Procrastination -.7532 .2962 -.1872 Studying .7202 .0172 -.2732 Time management .5962 .2452 -.1302
.249 .184 .178
Proportion of Variance For Each Factor: Eigenvalues
proportion of variance
accounted for
Eigenvalue / # of variables Note: here we have divided each eigenvalue by 9 (number of variables)
• communalities tell us proportion of variance in each variable that is explained by all of the factors you’ve included in your model
• so, there will be a communality for each variable in the analysis
• to calculate the communality for each factor: • square each factor loading • sum up the factor loadings within each variable (rows)
Proportion of Variance For Each Variable: Communalities
Proportion of Variance For Each Variable: Communalities
Variables Factor 1 Factor 2 Factor 3
Assignment effort -.1572 .6292 .3882 Class attendance .2582 .7232 .1242 Office hours visits .0632 -.6632 .2062 Math love -.1222 -.0912 .8642 Previous experience .3722 .3702 .6692 Partying -.7422 .0382 -.2722 Procrastination -.7532 .2962 -.1872 Studying .7202 .0172 -.2732 Time management .5962 .2452 -.1302
squaring each factor loading on original output (same as for eigenvalue calculations)…
Proportion of Variance For Each Variable: Communalities
Variables Factor 1 Factor 2 Factor 3 h2
Assignment effort -.1572 .6292 .3882 Σ = .571 Class attendance .2582 .7232 .1242 Σ = .605 Office hours visits .0632 -.6632 .2062 Σ = .486 Math love -.1222 -.0912 .8642 Σ = .770 Previous experience .3722 .3702 .6692 Σ = .723 Partying -.7422 .0382 -.2722 Σ = .626 Procrastination -.7532 .2962 -.1872 Σ = .690 Studying .7202 .0172 -.2732 Σ = .593 Time management .5962 .2452 -.1302 Σ = .432
• interpretation of communalities (example using first variable):
(a) 57.1% of the variance in assignment effort is accounted for by the three overarching factors combined (Exam Prep, Coursework, Experience)
…or…
(b) the three factors explain 57.1% of the variance in assignment effort
Proportion of Variance For Each Variable: Communalities
Proportion of Variance For Each Variable: Communalities
When we have low communalities: o factors don’t account for much variance in that variable
…or… o the variable doesn’t have much in common with the other variables in the analysis
Possible causes of low communalities:
1) the variable is actually very different from the other variables 2) the measurement of that variable was unreliable 3) an insufficient number of factors was extracted
o extract only those factors that have eigenvalues greater than 1
o why greater than 1?
when eigenvalue > 1: a factor has loadings from 2 or more variables
if a factor has loadings from only one variable (so, that variable is its own factor), its eigenvalue will be about 1
if the eigenvalue of a factor is greater than one, it describes more variance than one variable could alone
Deciding on Number of Factors Method #1: Eigenvalue > 1
1) output a scree plot via statistical software (SPSS) 2) look to see where data elbows 3) number of dots above the curve reveal number of factors to extract
Deciding on Number of Factors Method #2: Scree Plot
this graph suggests extracting three factors from our data
• mathematical technique that provides a simpler description of the relationships among variables
Varimax Method of Rotation o orthogonal type of rotation (keeps the axes at 90°)
o forces the factor loadings to get closer to ‘0’ or ‘1’ (remember factor loadings are the correlations of the variables with the factors)
o essentially, with this rotation, each variable will strongly load on only one factor (and not so much on the other factors)
o this way, you don’t have any “medium sized” factor loadings, and it will be easier to interpret which variables belong to which factors
o resultant factors will also be independent (uncorrelated) because they will be defined by unique variables
Rotation
Factor Analysis: Example
Nine variables potentially contributing to final exam grade in Psych 3800…
each row represents a given participant’s score on each of the nine measures
SPSS Example: The Data
Analyze Dimension Reduction Factor…
SPSS Example: Factor Analysis
move all of the variables that you would like to subject to
factor analysis into the “Variables” section
SPSS Example: Factor Analysis Extraction Menu
stick with default method of extraction
Let SPSS decide on optimal number of factors using the
“eigenvalue > 1” method
request scree plot
SPSS Example: Factor Analysis Rotation Menu
request Varimax (orthogonal) rotation
SPSS Example: Factor Analysis Scores Menu
request that the factors be saved in your current data file
for further analysis
Note: these outputted values are helpful for examining correlations between extracted factors
communalities for the extracted factors are provided by SPSS
SPSS Output: Factor Analysis Communalities
Interpretation (example): the extracted factors explain 60.5% of the variance in class attendance
SPSS Output: Factor Analysis Total Variance Explained
eigenvalues for three extracted factors (unrotated)
proportion of variance accounted for by each of the three extracted factors
(unrotated)
so, the first factor in the unrotated solution has an eigenvalue of 2.370 and accounts for 26.337% of the variance in the data
SPSS Output: Factor Analysis Total Variance Explained
eigenvalues for three extracted factors (rotated)
proportion of variance accounted for by each of the three extracted factors
(rotated)
so, the first factor in the rotated solution has an eigenvalue of 2.238 and accounts for 24.865% of the variance in the data
SPSS Output: Factor Analysis Scree Plot • scree plot seems to suggest that a three-factor solution would be ideal • in this case, the scree-plot method and the eigenvalue > 1 method are in agreement (this will not always be the case)
three-factor solution noted (but difficult to interpret without subsequent rotation)
SPSS Output: Factor Analysis Component Matrix This table provides the factor loadings for all of the variables on the factors
extracted using the eigenvalue > 1 method, prior to rotation.
SPSS Output: Factor Analysis Rotated Component Matrix This table provides the factor loadings for all of the variables on the factors
extracted using the eigenvalue > 1 method, after rotation.
three-factor solution is much easier to interpret now that the factors are made to be independent
SPSS Output: Factor Analysis Rotated Component Matrix
The first factor has its highest loadings from the highlighted variables…
these variables seem to suggest an exam preparation factor (what you do with your time in the weeks leading up to the exam)
Factor 1 is made up of low partying and procrastination (negative loadings) and high time management and studying (positive loadings).
SPSS Output: Factor Analysis Rotated Component Matrix
The second factor has its highest loadings from the highlighted variables…
these variables seem to suggest a coursework factor (effort put into the various components of the course)
Factor 2 is made up of low office hours visits (negative loadings) and high class attendance and effort put into assignments (positive loadings).
SPSS Output: Factor Analysis Rotated Component Matrix
The third factor has its highest loadings from the highlighted variables…
these variables seem to suggest a math experience factor (generalized familiarity of and appreciation for math)
Factor 3 is made up of high appreciation for math and greater experience with stats (positive loadings)
regression method was used to output scores on each factor for every participant (can be used in subsequent analyses of the three factors)
SPSS Output: New Variables
Final Assignment (yay!)
• not written as an APA-style results section • answer all questions fully, in numbered format
• maximum of 2.5 pages, double-spaced • APA formatting required
• answer all questions in sentence-form (no point-form this week)
Assignment 10: Overview
• submit all output as well
Question Tips and Hints
• for question #2 and question #5: argue both sides of each issue provide statistics or evidence from your output to support each side presented
• you will run three separate analyses in order to be able to answer the assignment questions:
(a) factor analysis (principal components, Varimax rotation) (b) bivariate correlation of all variables (c) bivariate correlation of factor scores
• all remaining questions are self-explanatory (answer all parts of each question)
• submit the factor analysis assignment to me by Friday, April 11, no later than 9:00AM
o can send it via e-mail ([email protected]) o can submit to me in person in my office o can slide it under my door if I am not in my office, but please send me an e-mail letting me know it’s there
Assignment Submission
• no further labs
• office hours next week Tuesday, April 8, 9:30AM-11:30AM o after that, set up appointment if you would like to meet o e-mail me your questions o note: I do not get to see the exam, so exam-related questions should be directed to Dr. McRae
• I’ll be in touch with an update of how to pick up your remaining assignments
• don’t forget that practice labs are available on Sakai/OWL for you to check out if you would like extra practice
all units should be posted by next week answers are available for all practice labs
What’s Next?
Good luck on the exam! Thank you!