plotting correct values for gam model in r

Question

I have the following example (which is in similar format to my data):

set.seed(1)
RandData <- runif(8760*2)
Locations <- rep(c('UK','France'),each=8760)

Date = seq(from=as.POSIXct("1991-01-01 00:00"), 
           to=as.POSIXct("1991-12-31 23:00"), length=8760)

Final <- data.frame(Loc = Locations,
                    Doy = as.numeric(format(Date,format = "%j")),
                    Tod = as.numeric(format(Date,format = "%H")),
                    Temp = RandData)
require(mgcv)
mod1 <- gam(Temp ~ Loc + s(Doy) + s(Doy,by = Loc) +
  s(Tod) + s(Tod,by = Loc),data = Final)
plot(mod1,pages = 1, scale = 0)

From this model, the outcome which is plotted firstly shows the mean temperature variation as a function of day of year and then the amount that the temperature at each location varies from this mean. The the same is shown for the Time of day.

The problem with this is that the yaxis do not show the correct range of values i.e. the first plot does not show the mean temperature for these locations. How can the plots be altered to show (1) the mean temperature for both locations and (2) the temperature at each location compared to the mean instead of some arbitrary number.

If I haven't been clear in expressing my intentions let me know and I will try to provide a better example.

The plot uses the correct values. Your interpretation of the plots is wrong. — Thierry, Jun 27 '12 at 11:49

Thierry · Accepted Answer · 2012-06-27T13:51:16.827

0

The plots of a gam model display the effect of the smoother itself. If you want the combined effect of several part of the model, then create a data.frame with the relevant values for each value, calculcate prediction for those values with predict(model, newdata) and create the plot you want.

dataset <- expand.grid(Loc = unique(Final$Loc), 
    Doy = pretty(Final$Doy), Tod = pretty(Final$Tod))
dataset$Temp <- predict(mod1, newdata = dataset)
library(ggplot2)
ggplot(dataset, aes(x = Doy, y = Temp, linetype = Loc)) + geom_line() + 
    facet_wrap(~Tod)

edited Jun 27 '12 at 13:51

answered Jun 27 '12 at 11:48

Thierry

18,049
5
48
66

could you provide an example from the question posted above? I understand what you are saying, just find it difficult to apply to a gam with two locations (i.e. UK and France)# – KatyB Jun 27 '12 at 12:44
Could you please specify why the example is not reproducible. Also I think that the script provided may work, I think the problem lies with the plotting, the ggplot returns: Error in aes(x = as.factor(Doy), y = as.numeric(Temp)) + geom_line() : non-numeric argument to binary operator – KatyB Jun 27 '12 at 13:33
Loc must be a factor to get your code running on my machine. My code was missing some brackets (corrected now). – Thierry Jun 27 '12 at 13:52

plotting correct values for gam model in r

1 Answers1