justinpriest
commented
5 years ago Hi Sara,
I'm unclear how the code above is different than what I did. I run the code, and it looks the same as the output from what I did, just with color. I purposefully did not add color because I was trying to make it publication worthy and only grayscale. If I add lines to this at the 0.005 and 0.995 probabilities, I get the same results as before:
You could also add the following line to the ggplot that you've produced above
geom_contour(aes(x=Q9abs, y=Q2, z=prob), color="black", breaks = c(0.005, 0.1, 0.5, 0.9, 0.995))
As you can see, it makes lines that are parallel to the original break. As I understand it (and I'm still kind of struggling to make exact sense of it all), since this is the "response" value, it is the probability of being a certain age (i.e., probability of being age-2). And since it is a probability (not error or confidence intervals), the contours of these would be parallel to the 50% probability.
The marginal plots from above only show one variable at a time. In my opinion, we would still want something similar to the Figure 4 plots that I've created, with both variables as the axes, and the actual ages plotted (correct/incorrect).
What do you think?
fssem1
https://cran.r-project.org/web/packages/sjPlot/vignettes/plot_marginal_effects.html
Instead of the plot 'model_pred_binom' I think you want a marginal model plot of the different variables. Or the function marginalmodel plots here: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4885900/ Another example: coho_scales_fulldata <- coho_scales_fulldata %>% filter(Location != "AL") coho_scales_fulldata <- coho_scales_fulldata %>% filter(Location == "HS")
binomfit <- glm((Age-1)~Q9abs+Q2, data=coho_scales_fulldata, family="binomial") newdata <- expand.grid(Q9abs=seq(min(coho_scales_bothriv$Q9abs), max(coho_scales_bothriv$Q9abs), length.out = 100), Q2=seq(min(coho_scales_bothriv$Q2), max(coho_scales_bothriv$Q2), length.out = 100))
newdata$prob = predict(binomfit, newdata, type="response")
ggplot(newdata, aes(Q9abs, Q2, fill=prob)) + geom_tile() + scale_fill_gradient2(low="red",mid="yellow",high="blue",midpoint=0.5, limits=c(0,1))