Transcript Slide 1
How to calculate:
- Average Effect Size (step 4)
- Moderators (step 5)
Overview of PPT slides
We will first go through “how” to do this.
Then, we will repeat everything and explain
“why” we do each step
Overview of Data Analysis
Central tendency – Effect size
Variability – Homogeneity test
What is the average effect and is it significant?
Does average affect have variability?
(If so, then you test moderators)
Prediction
Does the average effect differ with moderators?
Average effect size (ES)
Overview
Conceptually…
First, transform “r” into Fisher’s z-to-r
Second, weight them by sample size/inverse variance
Third, sum them together
Fourth, divide by sum of total sample size.
In Practice…
You transform “r” into Fisher’s z-to-r
You calculate inverse variance
Then use Macro to do the rest
At this point you have:
(1) Effect sizes. (2) Sample sizes. (3) moderators
Step 1 - Transform “r” into Fisher’s z-to-r
See formulas from “Example-DataSet2”
Step 2 - Weight by inverse variance
See formulas from “Example-DataSet2”
Step 3 - Upload to SPSS
in SPSS, file-open
Step 4 - Initiate “MeanES” macro
Download from Lipsey/Wilson website
Open YOUR DATASET
Open a new syntax
Put the following at the end of the syntax file
INCLUDE 'C:\Documents and Settings\Desktop\metaclub\MeanES.SPS' .
Highlight the sentence
Click run (blue triangle at top)
Step 5 - Calculate “MeanES”
In syntax, type the following sentence:
MEANES ES = ES_zr /W = weight /PRINT IVZR.
fyi – “ES_zr”
is my name for the effect sizes
“weight”
is my name for the weight
“/PRINT IVZR” converts output back to “r”
Highlight and click run
Step 6 – Interpret output
ES = .1225, p = .0000
Q = 1140.11, p = .0000
Now, let’s repeat and explain “why”
Transform into Fisher’s z-to-r
Weight by inverse variance
Why? “r” does not have a normal distribution because it is
skewed at the tails. Fishers’ z-to-r has a normal distribution, so it
is the preferred metric.
Why? The larger the sample size of a particular study, the larger
the impact. Weighting by the inverse takes into account sample
size.
Control for standard error
Why? When you weight by the inverse variance, then the
standard errors are incorrect, so you can’t use the traditional
tests within SPSS. The macros by Lipsey/Wilson control for the
problem with standard error.
Now, let’s move to Moderators
Overview
Conceptually…
First, need to ascertain “homogeneity” which tells you if
variance exists in average effect size
Can test categorical moderators (categories like college
student versus actual juror) similar to ANOVA
Can test continuous moderators (such as length of
stimulus) similar to Regression
In Practice…
Use macros
Macros exist for ANOVA & Regression
Macros for moderators
METAF – Categorical Moderators
INCLUDE 'C:\Documents and Settings\Desktop\metaclub\MetaF.SPS' .
METAF ES = ES_zr /W = weight /GROUP = ev1_subjecttype /PRINT IVZR.
METAF – Continuous Moderators
INCLUDE 'C:\Documents and Settings\Desktop\metaclub\MetaREG.SPS' .
METAREG ES = ES_zr /W = weight /IVS = ev1_subjecttype /PRINT IVZR.
fyi – can only run 2 macros in same session
fyi – I run categorical moderators using both METAF and
METAREG because answer different questions
Interpreting Categorical Macro
Sig difference = 22.87, p = .0000
Interpreting Continuous Macro
beta = -.0958, p = .0014
Now, to Multivariate
METAREG can handle multiple variables
METAREG ES = ES_zr /W = weight /IVS = ev1_subjecttype ev2_stimulustype /PRINT IVZR.
fyi – Can include as many variables as you wish.
fyi - I believe you can include CATEGORICAL
moderators IF:
(1) they are dichotomous,
(2) they are continuous and linear relationship
Interpreting Continuous Macro
beta for each one, p value for each one
overall r-squared
Finally, Interaction analysis
Center each variable
Create interaction term by multiplying together
Enter all three into METAREG
If interaction term is sig, then interaction exists
How to graph the interaction? (next week)
FYI
Our website has my excel file (Example-DataSet2) and
the accompanying SPSS file (SPSS-ExampleDataSet2)
You also have my quals paper, so you can use the SPSS
file to practice and see if your data match the quals
paper.
HOWEVER, the data will only match the CATEGORICAL
moderator analysis (not the continuous moderator
analysis) for reasons I don’t have time to go into.