I have a regression model with a significant interaction term between group a and b. In one single graph, I would like to plot the regression line of the whole group as well as the residuals of group b in different colors (4 colors) depending on the distance to the line.
Does anyone know how I can do this, and whether there is a specific package that would work best for this?
Does you interaction involve one dichoutomous categorical predictor and one continuous predictor? And your effect implies different slopes for the two groups?
And do you want to have a scatterplot with an overall regression line, despite the interaction?
And another question: do you want a simple scatterplot, but the points in a different color depending on their absolute value of the corresponding residual?
Dear Boris, thanks for replying.
Yes, there is a difference in the association of two continuous variables between a binary categorical predictor.
Yes, I would like a scatterplot with an overall regression line despite the interaction (I want to show that as compared to the total population, a specific subgroup is different and also to what extent this subgroup deviates from the mean).
Yes, I would like a simple scatterplot, however, I would like it with a different color for the residuals based on the standardized residuals (four groups, split up by SD).
Dear Tim,
I am not completely sure if this here is what you need (but I hope it is):
# simulate data
N
Of course you may use other logical expressions to get different colors depending on other values defining a particular group or you may combine variables for group membership and the concrete values of the residuals.
There is also (very likely) a shorter way (less lines of code) to get the graph above. But this way it is hopefully easier to understand how it works.
Dear Boris, this is definitely getting me closer to what I was trying to achieve, thanks! I have been able to plot my own data using the syntax. However, what I would like now is to be able to select a subset of people from the whole dataset, and plot mod with the residuals of only that subset. I have been trying to do this myself, but haven't gotten it to work. Do you have any idea on how to do this?
Dear Tim, I am glad that the code was helpful.
Okay, so assuming you want to get the scatterplot only for a subset of your data, you could use the logical expressions for the variables in your data sets. Here is an example:
# simulate data
N
Definitely another step closer Boris, I have been trying to combine both of your syntaxes like this:
subset= gender==1 & abs(rstandard(mod))>=1&abs(rstandard(mod))
Dear Tim, I am glad that the code was helpful again. When you work with variables in a data.frame-object you need to either use the attach()-function to make the variables in the workspace available or use the $-operator:
# simple scatterplot without plotting the points yet
plot(data$x,data$y,type="n")
# plot points for gender == 0 and residuals between 1 and 2
points(data$y~data$x,subset= data$gender==1 & abs(rstandard(mod))>=1&abs(rstandard(mod))
Dear Boris, first of all thanks so much for helping out!
I was able to use your code. I had to remove the abs() part to actually get the minus values correctly extracted from the model. I stratified the subsets to the 4 I had in mind and also added in a grey confidence interval for the line. Attached is the final product.
For posterity, it is also quite simple to do this with ggplot2:
## using the above simulation data
N
Dear Matthew, thanks for the ggplot2 syntax. Quite similar although in ggplot2 it is simpeler to add lines etc. to the plot.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R Statistical Package