I have a GAM with country now added as a factor, to model the relationship within countries (i think this has been done correctly, let me know if not), and now wish to plot my new model how ever my issue is an aesthetics error (seen below), if anyone knows how to correctly plot this that would be great

! Aesthetics must be either length 1 or the same as the data (480): y

This is the code i have used and a snipet of my data (let me know if you need my whole dataset and i can attach it), hopefully someone is able to plot this, i think the error is coming as im not plotting the model correctly now i have added countries as a factor

model code;

mod = gam(gdp_per_capita ~ s(fisheries_production_pc, k = 25, bs = 'cs')   as.factor(Country_Name), 
           data = economy_df, 
           family = gaussian(link = "log"))

#predictions
preds = predict(mod, type = 'response', se.fit = TRUE)

#plot model
plot = ggplot(economy_df, aes(y = gdp_per_capita, x = fisheries_production_pc))  
geom_point()   
geom_line(aes(fisheries_production_pc, preds$fit), colour = 'red')  
scale_x_log10()

data;

economy_df

Country_Name year gdp_per_capita fisheries_production
Albania 1997       717.3800              1110.80
Albania 1998       813.7894              2807.50
Albania 1999      1033.2425              3057.90
Albania 2000      1126.6833              3635.00
Albania 2001      1281.6598              3597.20
Albania 2002      1425.1242              4516.80
Bosnia  1997       982.8018               253.00
Bosnia  1998      1102.3907               254.00
Bosnia  1999      1251.7476               255.00
Bosnia  2000      1484.1761               255.00
Bosnia  2001      1544.6021               255.00
Croatia 1997      5312.3695             20551.49
Croatia 1998      5691.1095             27935.08
Croatia 1999      5246.9360             25222.19
Croatia 2000      4887.7137             27944.24
Croatia 2001      5412.9251             29019.12
Cyprus 1997     14234.2441             25788.00
Cyprus 1998     15092.8262             20482.00
Cyprus 1999     15287.9189             41060.00
Cyprus 2000     14388.3477             70223.00



  

CodePudding user response：

I think you can do something like this:

ggplot(economy_df, aes(y = gdp_per_capita,
                       x = fisheries_production_pc,
                       color=Country_Name))  
  geom_point()   
  facet_wrap(~Country_Name, scales = "free")  
  geom_smooth(method="gam", formula=y~s(x,bs="cs"))  
  scale_x_log10()

CodePudding user response：

It is somewhat easier to use the emmeans package to generate predictions, especially if you want to average (marginalize) over non-focal predictors (e.g. generating a country-independent prediction from a model that has country as a fixed effect); you can also use the ggpredict package for this (which wraps the emmeans package)

library(mgcv)
library(emmeans)
library(ggplot2)
mod <- gam(gdp_per_capita ~ s(fisheries_production_pc)   factor(Country_Name), 
          data = economy_df, 
          family = gaussian(link = "log"))

## predictions
fpcvec <- exp(seq(log(200), log(40000), length.out = 51))
predfun <- function(mod, specs) {
    as.data.frame(
        emmeans(mod,
                specs = specs,
                data = economy_df,
                type = "response",
                at = list(fisheries_production_pc = fpcvec))
    )
}
pred1 <- predfun(mod, specs = ~fisheries_production_pc)
pred2 <- predfun(mod, specs = ~fisheries_production_pc*Country_Name)

#plot model
plot <- ggplot(economy_df,
              aes(y = gdp_per_capita, x = fisheries_production_pc))  
    geom_point()  
    facet_wrap(~Country_Name)  
    geom_line(data = pred1, aes(y = response))  
    geom_line(data = pred2, aes(y = response), colour = 'red')  
    scale_x_log10()
ggsave("fishpred.png")

Notes:

• here I'm drawing both the country-independent (marginal) prediction (in black) and country-specific prediction (in red)
• if you want to add confidence regions you can use geom_ribbon with the lower.CL and upper.CL variables that emmeans generates
• since you fitted an additive effect of country, the country-specific predictions are all parallel (on the log scale!) to the overall prediction/each other. (A model with level-specific smooths would have broken with this small data set ...)
• Pedersen et al 2019 (ref below) is a good reference for the choices you should think about when fitting this kind of multi-country GAM.

Pedersen, Eric J., David L. Miller, Gavin L. Simpson, and Noam Ross. 2019. “Hierarchical Generalized Additive Models in Ecology: An Introduction with Mgcv.” PeerJ 7 (May): e6876. https://doi.org/10.7717/peerj.6876.