TO: Students in CJ 405
RE: Homework for 4/22
DATE: 4/19
PURPOSE. The purpose of this homework is to run a regression that first puts in two main effects, then adds an interaction effect. We are working with the individual level data file.
The dependent variable is Q31: how strongly you believe the police can protect you from violent crime. Look up the variable so you are clear how it is coded.
You first explore two main effects: living in Philadelphia (0=no;1=yes) and being nonwhite (0=white; 1=nonwhite). You will need to create this variable NONWHITE as part of your run; compute statements are shown below.
We would expect a main effect of each of these two variables: you will tell me why in your homework.
But then we are also going to look at the interaction of the two. You are also going to need to create the interaction term. I call this variable PHILNONW. Given how the main effects of each portion of the interaction term are coded, what CELL of the interaction term are we saying is of central interest? (HINT: what cell is coded "1" on both of the main effects?) Do you think that people in this group will have MORE or LESS confidence in police's ability to protect you from violent crime? Why?
Run the regression below. You may need to type this in because some
of these options (CHA on the STATISTICS line for example) I cannot
generate with a menu. IT IS EXCEEDINGLY IMPORTANT THAT WHAT YOU SEE
ON THE
STATISTICS= LINE
SUBCOMMAND FOR THE REGRESSION COMMMAND APPEAR ON YOUR COMMAND SYNTAX.
IT IS ALSO IMPORTANT THAT YOU ENTER THE PREDICTORS IN TWO BLOCKS AS
SHOWN ON THE
METHOD=
SUBCOMMAND LINE
Get the mean scores on the outcome for each cell in the interaction term. You can use the compare means command. We demo-ed this in class last week.
The regression command does two regressions: first, one with both main effects. Second, another regression with the two main effects and the interaction term added.
Of crucial importance here is the F test for R squared change. When you add in the interaction term you change the R squared, and there is an F test of that. The conceptual logic of that is explained in depth in your latest note file.
HINT for RUNNING:
* I have run this syntax file BOTH on SPSSW 6.1 and 8.0 so it should
work for you. Remember, you may need to be sure the subcommand lines
are indented.
* You will need to tell it where to look for your data file, and this
location will be different from the location I use.
WHAT I WANT
1. Looking at the regression with just PHILLY and NONWHITE, interpret the b weights, t tests of b weights, and R squared. (5 points)
2. Looking at the regression with PHILLY, NONWHITE, and PHILNONW, interpret the R squared change (from the above regression), and the F test of the R squared change. Use an alpha=.15 for the F test and an F critical for the F test of R squared change of 2.2. DO NOT try and interpret any of the b weights in this regression with these three predictors. (3 points)
3. Interpret what the interaction term is telling you by looking either at a scattergram (see below, for example), of the means in each cell of the crosstab PHILLY*NONWHITE. (1 point)
4. Conceptually, tell me what you think is happening here, and why. (1 point)
GET
FILE='D:\PCW\UMETH\survey\UMET99FF.sav'.
EXECUTE .
COMPUTE NONWHITE = 0 .
EXECUTE .
IF (WHITE=0) NONWHITE = 1 .
EXECUTE .
COMPUTE PHILNONW=0 .
EXECUTE .
IF ((NONWHITE=1) AND (PHILLY=1)) PHILNONW=1 .
EXECUTE .
SAVE OUTFILE='D:\PCW\UMETH\survey\UMET99GG.sav'
/COMPRESSED.
GRAPH
/SCATTERPLOT(BIVAR)=philly WITH q31 BY nonwhite
/MISSING=LISTWISE .
REGRESSION
/DESCRIPTIVES MEAN STDDEV CORR SIG N
/MISSING MEANSUB
/STATISTICS COEFF OUTS R ANOVA LINE COLLIN TOL CHA
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT q31
/METHOD=ENTER philly nonwhite /METHOD=ENTER philnonw .