Visualizing an interaction effect of a Logistic Regression Analysis (in SPSS) What would be the best way to go about visualizing an interaction effect of a Logistic Regression Analysis (preferably using SPSS)?
There is a significant interaction between one interval (HDI) and one binary (before/after Fukushima) variable as well as between that same interval variable (HDI) and another interval variable (share of non-renewable electricity consumption). The dependent variable is binary (ecological benefits of solar power mentioned in news article about topic yes/no).
Failing any kind of more specific way of doing this, what would be more general steps to arrive at some visualization? Is a visualization even a good idea? (For context: Until now I have been looking at interactions using ANOVA where plotting interaction effects is straightforward. Those visualizations have been tremendously helpful and I would like something similar for this.)
 A: Visualization is certainly a good idea if these findings are important to communicate.  A narrative description of an interaction effect may be cumbersome to write and to absorb, and it is unlikely to make the same impact that a chart would.  
I'd make a bubble chart.   This is not entirely an SPSS solution, but if you use Excel or R it will work, especially for the continuous-continuous interaction and especially if you are not concerned about making the equivalent of partial plots, i.e., if you don't need to show how the dependent variable is a function of these independent variables while controlling for others.*
Start with a grid consisting of values of HDI on one axis and share of non-renewable electricity consumption on the other.  Divide each predictor into some discrete regions (5? 10? 50?  It depends on your judgment and your facility with SPSS recodes, or with SPSS's auto-recode commands). For each X-Y region, plot the bubble size as the mean of your dependent variable in that region, taking Yes to be 1 and No to be 0.  
If you use R, instead of bubble size you have the option of varying the points using color or symbol.
The continuous-binary interaction could be done the same way; it'll just look a little simpler, perhaps a little simplistic.
*If you do want to incorporate such control, first regress Y on the control variables--those you're not interested in plotting.  Then, instead of using the mean of Y as your plot variable, use the mean of the residuals from that regression.  The tricky thing here will be what to express about the range of values for these residuals, since they won't be bounded by 0 and 1.
If I've left out some important step someone will correct me....
EDIT:  you could make this entirely an SPSS operation if you discretized your Y variable and created a scatterplot of X1 with X2...a) using the "by" command to plot Y in multiple colors, or b) using the "by [Y] (identify) command to plot Y in multiple characters.
