interact_plot point color

[jacob-long]

interact_plot point color

[jacob-long]

It should be doing what you want it to, so something must be going wrong. Here's an example of the intended output:

I'll give a note on how the coloring of plotted points for continuous moderators works. Since ggplot2 won't let me color the lines with three different colors (a discrete scale) plus points with a continuous scale, I use a little trick. I take the darkest of the moderator colors and then scale the points' alpha, making them more transparent when the moderator is low.

With that in mind, one possibility other than a true bug that could make all the points appear to be the same color is if moderator is highly skewed.

[J0PR]

I noticed this when changing the color to something other than "blues"; with that scale things look OK. This happens because the color of the points was set by taking the color of the "+ 1 SD" and then changing the alpha values, which is close to what that colorscale does. Here's how it looks with another colorscale:

As you can see, the points are just purple, with a varying degree of alpha values.

I edited the "else if" on line 429 to:

    else if (!is.factor(d[, modx])) {
      d$sd_split[d[[modx]] > mean(d[[modx]]) + sd(d[[modx]])] <- colors[1]
      d$sd_split[d[[modx]] < mean(d[[modx]]) + sd(d[[modx]])] <- colors[2]
      d$sd_split[d[[modx]] < mean(d[[modx]]) - sd(d[[modx]])] <- colors[3]
      p <- p + geom_point(data = d, aes_string(x = pred, 
        y = resp, size = "the_weights"), colour = d$sd_split, 
        inherit.aes = FALSE, show.legend = FALSE)
    }

This solved the problem for my particular plot (also interaction with 2 scale variables).

[jacob-long]

That's a really interesting solution to this issue. As you noticed, the "trick" I use with continuous moderators is to use the alpha aesthetic. The predicted values of the moderator are discrete and the current version of ggplot2 won't allow me to mix discrete and continuous color scales (there is some discussion of allowing that in the future: tidyverse/ggplot2#2570).

Of course, my trick only works if one of the continuous color scales is used (e.g., "Blues", "Reds", "Greys") and looks odd like you found if something like "PuRd" is used. I'll give some thought to how cleanly I can detect the use of the other color scales and how they look with this scheme.

[MikeBurkePhD]

I noticed this when changing the color to something other than "blues"; with that scale things look OK. This happens because the color of the points was set by taking the color of the "+ 1 SD" and then changing the alpha values, which is close to what that colorscale does. Here's how it looks with another colorscale:

As you can see, the points are just purple, with a varying degree of alpha values.

I edited the "else if" on line 429 to:

    else if (!is.factor(d[, modx])) {
      d$sd_split[d[[modx]] > mean(d[[modx]]) + sd(d[[modx]])] <- colors[1]
      d$sd_split[d[[modx]] < mean(d[[modx]]) + sd(d[[modx]])] <- colors[2]
      d$sd_split[d[[modx]] < mean(d[[modx]]) - sd(d[[modx]])] <- colors[3]
      p <- p + geom_point(data = d, aes_string(x = pred, 
        y = resp, size = "the_weights"), colour = d$sd_split, 
        inherit.aes = FALSE, show.legend = FALSE)
    }

This solved the problem for my particular plot (also interaction with 2 scale variables).

I am having this same exact issue. @J0PR where exactly did you edit the code? I am hoping your solution works for me too.