**David A. Kenny**

**September 15, 2014**

** **

**GAPIM-I Macro **

**This
macro, called GAPIM_I, was written by David A. Kenny, Department of Psychology,
**

**Thank You!**

**I
thank Randi Garcia who assisted on this project and provided me example data. **

**Data Preparation**

**The
dataset needs to be in individual format: One record for each group member.
There need to be at least three members per group who do not have any missing on X (the
group composition variable) and Y (the outcome) and ideally there are at least 20 groups.. The current version of the
program presumes that X is a dichotomy that is already effects coded (codes of
+1 and -1).**

** **

**Downloads:**

**GAPIM_I.sps (You need SPSS to
open this and the next 3 files.) **

**Macro Output (You do not
need SPSS to open this file.) **

**
**

**
****
**

**
The text file is NOT contained in the SPSS
output file. It is contained in a file called “c:\gapim_i.txt”. Look
for it there!
**

**
**

**The Macro Call**

** **

**
This
is the statement for the sample data:**

** GAPIM_I x = gender/y = gidscale /groupid = groupnum xn = 'Gender' yn = 'Group Identification' **

groupn="group" upper_lab="Women" lower_lab="Men".

**The
defaults are as follows:
**

**x = X **

**y = Outcome **

**groupid****
= group
**

**
**

**malpha = .10
**

**upper_lab=Highs**

**lower_lab=Lows**

**oflile ****= c:\gapim_i.txt**

**directory****
= c:\
**

**groupn ****= group**

**xmiss****
= 1
**

**clist****
**

**That
is, if you just say “GAPIM_I.”, the program will assume that are variables in
the SPSS data file with variables named X and Outcome. The variables in the macro are as follows:
**

** x = name of the group composition variable in the SPSS dataset, assumed to be a dichotomy and effects coded
**

** y = name of the causal variable
for the partner in the SPSS data set
**

** groupid = variable name of groups or team
**

** alpha = value used in significance testing**

** malpha = value used for marginal significance**

** upper_lab = label for the 1's on the X variable **

** lower_lab = label for the -1's on the X variable **

** ofile = name of the text file which GAPIM_I writes**

** directory = where output file and temporary files are written **

** nmiss **

** 1: delete group if one or more person has a missing value on X**

** 0: delete only the individual if a missing case on X**

** groupn = name for groups (e.g., class or team)**

** clist = name of covariates in the SPSS dataset separated by spaces**

**
GAPIM_I was written on SPSS 18 and there is
no guarantee that it work on earlier or later versions of SPSS. GAPIM_I creates a dataset with all of the GAPIM-I variables (X', I, I', contrast and diversity measurs), the name of which is gapimdata.sav which is written in directory designated by the user.
**

** **

**
Check frequently for updates, as there are
likely to be changes. No guarantee for accuracy. Almost certainly
you will need to edit the DataToText output in research reports. There
will be updates. There is no guarantee for accuracy. Examine not
only DataToText output file, but also the SPSS output file. The user
needs to carefully edit the GAPIM_I output in
research reports. Please cite this webpage if you do use it. Moreover, you need
a footnote in your paper that says: "Some of the material here was produced by the SPSS macro
GAPIM_I (Kenny, 2014)." Material lifted verbatim must be in quotes to avoid plagarism.
**

**Warnings**

**
**

**
**

**The maximum value of the X variable is
greater than one which is not ordinarily allowable.**

**2. ****The minimum value of the X variable is less
than minus one which is not ordinarily allowable.**

**3. ****The outcome variable is a dichotomy while
the analysis assumes it is not.**

**4. ****There are one or more missing observations
for the outcome, but groups containing those missing cases are still retained
in the analysis.**

**5. ****Because group sizes vary, the chi square
difference tests involving diversity, contrast, and group effects are
approximate.**

**6. ****The X variable is skewed which leads to
co-linearity.**

**7. ****There is multicollinearity between the
following predictors.**

**8. ****There are less than 10 groups, probably too
few for a GAPIM analysis. **

**9. ****There are between 10 and 19 groups, probably
a marginal number for a GAPIM analysis. **

** **

**Macro Output **

**
**

**WARNING: 1. Because group
sizes vary, the chi square difference tests involving diversity, contrast, and
group effects are approximate.
**

**GROUP
ACTOR-PARTNER INTERDEPENDENCE MODEL**

** **The Group Actor-Partner Interdependence Model is estimated by this macro. The computer outputs are in the SPSS Output file. The data file with the transformed variables is located at c:\tempgapimdata.sav. The X variable in this dataset is Gender the others variable is X', the actor similarity variable is I, and the others similarity variable is I'. Also, the Xmean variable is the group mean (Norm), the Xcon variable in this dataset is X - X' (contrast), the XconI variables is I - I' (similarity contrast), and Xdiv is a measure of group diversity. The text file with a description of the results is located at c:\tempgapim_I.txt.

** **The group composition variable is Gender, a dichotomy consisting of Men and Women, and the outcome variable is Group Identification. The variable Gender is presumed to affect Group Identification in fourdifferent ways: the effect of the actor's own Gender or X, the effect of other group members' Gender or X', the ffect of the similarity of the actor's Gender to the other group members' Gender or I, and the effect of the similarity of the others' Gender or I'. All of these variables are effect coded. For instance, X is coded Men = +1 and Women = -1. Any group which contains missing data on Gender is dropped from the analysis.

**
**

**Descriptives**

** **For the analysis, there are 241 persons in 52 groups. The two group sizes are 4 and 5. There are 13 all-Women groups and 2 all-Men groups. There are 154 (63.9%) Women and 87 (36.1%) Men. The means and standard deviations of Gender and Group Identification are presented in Table 1.

**
**

**
**The effect of the actor's Gender is -.026 (p = .713). Men score .051 units lower on Group Identification than Women. The effect of the Gender of the other group members is .227 (p = .097). An actor, all of whose other group members are Men, scores .454 units higher on Group Identification than a member all of whose other members are Women. The effect of the actor's similarity to others in the group on Gender is .295 (p = .014). An actor, who is totally similar to the other group members on Gender, scores .589 units higher on Group Identification than an actor who is totally dissimilar. The effect of the others' similarity in Gender is -.210 (p = .106). In a 4-person group, a person with others who are completely homogeneous, either all Men or Women, scores .280 units lower on Group Identification than a person with 2 Men (or Women) and 1 Women (or Men) others.

** **The variance due to groups is equal to .128 and the proportion of variance due to group or intraclass correlation is equal to .135. The multiple correlation explained by the four GAPIM-I fixed effects is equal to .181. The chi square test with four degrees of freedom that compares the Complete Model to the Empty Model (a model with no predictors) equals 13.283 (p = .010). Because this chi square test is statistically significant, we can conclude that one or more terms of the Complete Model is needed. The chi square test with two degrees of freedom that compares the Complete Model to the Main Effects Model equals 8.652 (p = .070). Because this chi square test is not statistically significant, we do not have evidence that the interaction effects of the Complete Model are non-zero. The sample size adjusted BIC for the Complete Model is 665.79 whereas the value for the Empty Model is 675.83. Because the index is smaller, the Complete Model is a better fitting model than the Empty Model.

**The
Best Fitting Model
**

** **The best fitting model in terms of highest multiple correlation is the Interaction Contrast Model with a multiple R of .193. (The best fitting model in terms of lowest SABIC is also the Interaction Contrast Model with an SABIC of 665.224.) This implies that there is effect if the actor is different from the others in the group on Gender and the others are similar to each other on Gender. The chi square test comparing this model to the Empty Model is equal to 13.036 with 3 degrees of freedom (p = .005). The sample size adjusted BIC is equal to 665.224.

** Table 1: Descriptive Statistics**

**Variable Mean Standard Deviation**

**--------------------------------------------------------**

**Gender .278 .963**

**Group
Identification 3.910 .992
**

** Table 2: Model
Estimates and Fit**

**Model X X' I I' SABIC
R**

**-----------------------------------------------------------------------------------------------------------
**Empty

Main Effects

Complete

Constraints on both Main Effects and Interactions

** Table 3: Chi Square Tests**

**Model
Tested Chi Square df p
**

**------------------------------------------------------**

EMPTY MODEL AS THE COMPARISON MODEL (significant means a better fitting model)

Main Effects** ** ** **4.630 ** **2

Complete

Constraints on both Main Effects and Interactions

MAIN EFFECTS MODEL AS THE COMPARISON MODEL (significant means a worse fitting model)

COMPLETE MODEL AS THE COMPARISON MODEL (significant means a worse fitting model)

Constraints on both Main Effects and Interactions