Stats

From Technolotics

Contents 1 Factor Analysis 1.1 Exploratory Factor Analysis 1.2 Rotated Solution 2 Regression 2.1 Linear 2.2 Multiple 3 Transformations 3.1 Positively Skewed 3.2 Negatively Skewed 4 Trim Means 5 Case Deletion

Factor Analysis

Exploratory Factor Analysis

Analysis:

1) Analyse -> Data Reduction -> Factor Analysis
2) Move all IV’s into ‘Variables’ [nothing in selection variable]
3) Rotation: None
4) Descriptives: Univariate, Coefficients, KMO and Bartletts
5) Extraction: PCA, Unrotated, Scree, Eigenvalues over 1
6) Options: Exclude Cases pairwise, sorted by size, suppress values less than .3

Output:

1) Check KMO > .6
2) Check Bartlett Sig < .05
3) Check Scree Plot, how many points before elbow
4) Communalities – lowest extraction value = best explained by factors
5) Check Total Variance explained – which values with eigenvalues over 1 – how much cumulative variance do they explain?
6) Component matrix – items factors load onto which items

Rotated Solution

Analysis:

1) Analyse -> Data Reduction -> Factor Analysis
2) Rotation: Varimax (orthogonal) / Direct Oblimin (oblique)

Output:

1) Pattern Matrix - which items load onto which factors
2) What do factors mean?

Analysis:

3) Check reliability - Analyse - Scale - Reliability Analysis
4) Put items which load together in at once (i.e.: run once for each factor)

Output:

5) Alpha should be > .5 or .7

Question: How to do Confirmatory Factor Analysis? We wont have to

Regression

Linear

Analysis:

1) Analyse - Regression - Linear
2) DV = dependent
3) IV = independent
4) OK

Output:

1) Model Summary: R Square (*100) = percentage of varience explained by IV
2) Model Summary: Adjusted R Square = population varience explained by IV
4) ANOVA: F = ratio of varience explained by model : not explained
5) ANOVA: Sig = p of anova (<.05 for significance)
6) Coefficients: Unstandardized => B (Constant) = beta0 = constant
7) Coefficients: Unstandardized => B (IV) = beta1 = intercept
8) Regression Equation = beta0 + beta1 (x of IV)
9) Graphs - Scatterplot - Y Axis = DV, X Axis = IV

Question: Do we need to do independent samples t-test, if so how? analyse - compare means - independent samples grouping variable is IV2 => this must be dichotamous

Multiple

Question: Is this forced, standard or stepwise? This is standard.

Question: How to do others? select others from method drop down

Test for Suitability

1) Normality: Graph - Histogram
2) Linearity / Homoscedasticity / Residuals / Outliers: Graph - Scatterplot - Y Axis = DV, X Axis = IV
should be roughly in a line + clustered around center + rectangular distrobution + few outliers (3 per hundred)

Analysis

1) Analyse - Regression - Linear
2) Dependent = DV, Indepents = IVs
3) Stats: Estimates, Model Fit, R**2 Change, Descriptives, Part and Partial Correlations, Casewise Diagnostics - outliers outside 3
4) Plots: ZRESID = Y, ZPRED = X, Histogram, Normal

Output

1) Normal P-Plot - Are Red and Green Lines the Same?
2) Histogram - Normal?
3) Scatterplot - should be roughly in a line + clustered around center + rectangular distrobution + few outliers (3 per hundred)
4) Model Summary: R**2 (*100) = percentage variation in DV explained by both IV's
5) ANOVA: Sig = p of null hypothesis being true
6) Coefficients: B constant = constant = Beta0 | B (IV1, IV2 etc) = Beta1, Beta 2 etc
7) Regression Equation = Beta0 + Beta1(x of IV1 + Beta2 (x of IV2)
8) Coefficients: Standardized Beta = contribution of each IV
9) Coefficients: t sig = p of IV not making a unique contribution, should be <0.05
10) Coefficients: Correllation - Part (**2) = unique contribution of IV to DV

Heirarchical Multiple Regression

1) Repeat as above
2) Analyse - Regression - Linear => now in 'independents' box, click 'next'
3) Then add another IV
4) Run Analysis

Output

1) Model Summary - Change Statistics - R(**2) Change
= Additional amount of Variance explained by new IV (*100 = percentage)
2) Model Summary - Change Statistics: Sig F Change: should be <.05
3) Coefficients: Model 2 - See above for regression equation

Transformations

1) Transform - Compute: Target variable = new transformed variable
2) Insert veriable to be transformed into Numeric Expression box
3) Use ** to raise to a power, LG10 to Log, SQRT to square root
4) Graph -> Histogram - is a quick way to observe changes in variable

Question: How to transform using reciprocal?
not explained we dont need to know Question: Why is positively skewed to the left and not the opposite? chiara will check

Positively Skewed

1) Slopes to the left = positively skewed
2) mild: SQRT, mid: LOG, extreme: Reciprocal

Negatively Skewed

1) Slopes to the right = negatively skewed
2) Analyse - decriptives - frequencies - statistics - dispersion: maximum => read maximum value in output
3) Transform - Compute: (max - variable) + 1
4) Repeat Positive Step 2

Trim Means

Question: How?

Case Deletion

Question: How?