Skip to content

lattice::xyplot/ggplot2: plotting weighted data frames with lmline and smooth

4 messages · Dennis Murphy, Michael Friendly

Michael Friendly
# 
In the HistData package, I have a data frame, PearsonLee, containing 
observations on heights of parent and child, in weighted form:

library(HistData)

 > str(PearsonLee)
'data.frame':   746 obs. of  6 variables:
  $ child    : num  59.5 59.5 59.5 60.5 60.5 61.5 61.5 61.5 61.5 61.5 ...
  $ parent   : num  62.5 63.5 64.5 62.5 66.5 59.5 60.5 62.5 63.5 64.5 ...
  $ frequency: num  0.5 0.5 1 0.5 1 0.25 0.25 0.5 1 0.25 ...
  $ gp       : Factor w/ 4 levels "fd","fs","md",..: 2 2 2 2 2 2 2 2 2 2 ...
  $ par      : Factor w/ 2 levels "Father","Mother": 1 1 1 1 1 1 1 1 1 1 ...
  $ chl      : Factor w/ 2 levels "Daughter","Son": 2 2 2 2 2 2 2 2 2 2 ...

I want to make a 2x2 set of plots of child ~ parent | par+chl, with 
regression lines and loess smooths, that
incorporate weights=frequency.  The "frequencies" are not integers, so I 
can't simply expand the
data frame.

I'd also like to use different colors for the regression and smoothed lines.
Here's what I've tried using xyplot, all unsuccessful.  I suppose I 
could also use ggplot2, if I could do what
I want.

xyplot(child ~ parent|par+chl, data=PearsonLee, weights=frequency, 
type=c("p", "r", "smooth"))
xyplot(child ~ parent|par+chl, data=PearsonLee,  type=c("p", "r", "smooth"))

  panel.lmline  and panel.smooth don't have a weights= argument, though 
lm() and loess() do.

# Try to control line colors: unsuccessfully -- only one value of 
col.lin is used
xyplot(child ~ parent|par+chl, data=PearsonLee, type=c("p", "r", 
"smooth"), col.line=c("red", "blue"))

## try to use panel functions ... unsucessfully
xyplot(child ~ parent|par+chl, data=PearsonLee, type="p",
        panel = function(x, y, ...) {
            panel.xyplot(x, y, ...)
            panel.lmline(x, y, col="blue", ...)
            panel.smooth(x, y, col="red", ...)
            }
)

The following, using base graphics, illustrates the difference between 
the weighted and unweighted lines,
for the total data frame:

with(PearsonLee,
     {
     lim <- c(55,80)
     xv <- seq(55,80, .5)
     sunflowerplot(parent,child, number=frequency, xlim=lim, ylim=lim, 
seg.col="gray", size=.1)
     # unweighted
     abline(lm(child ~ parent), col="green", lwd=2)
     lines(xv, predict(loess(child ~ parent), data.frame(parent=xv)), 
col="green", lwd=2)
     # weighted
     abline(lm(child ~ parent, weights=frequency), col="blue", lwd=2)
     lines(xv, predict(loess(child ~ parent, weights=frequency), 
data.frame(parent=xv)), col="blue", lwd=2)
   })

thanks,
-Michael
Dennis Murphy
# 
Hi Michael:

Here's one way to get it from ggplot2. To avoid possible overplotting,
I jittered the points horizontally by +/- 0.2. I also reduced the point
size from the default 2 and increased the line thickness to 1.5 for
both fitted curves. In ggplot2, the term faceting is synonymous with
conditioning (by groups).

library('HistData')
library('ggplot2')
ggplot(PearsonLee, aes(x = parent, y = child)) +
   geom_point(size = 1.5, position = position_jitter(width = 0.2)) +
   geom_smooth(method = lm, aes(weights = PearsonLee$weight),
               colour = 'green', se = FALSE, size = 1.5) +
   geom_smooth(aes(weights = PearsonLee$weight),
               colour = 'red', se = FALSE, size = 1.5) +
   facet_grid(chl ~ par)

# If you prefer a legend, here's one take, pulling the legend inside
# to the upper left corner. This requires a bit more 'trickery', but
# the tricks are found in the ggplot2 book.

ggplot(PearsonLee, aes(x = parent, y = child)) +
   geom_point(size = 1.5, position = position_jitter(width = 0.2)) +
   geom_smooth(method = lm, aes(weights = PearsonLee$weight,
               colour = 'Linear'), se = FALSE, size = 1.5) +
   geom_smooth(aes(weights = PearsonLee$weight,
               colour = 'Loess'), se = FALSE, size = 1.5) +
   facet_grid(chl ~ par) +
   scale_colour_manual(breaks = c('Linear', 'Loess'),
                       values = c('green', 'red')) +
   opts(legend.position = c(0.14, 0.885),
        legend.background = theme_rect(fill = 'white'))


HTH,
Dennis
On Fri, Oct 21, 2011 at 8:22 AM, Michael Friendly <friendly at yorku.ca> wrote:
Michael Friendly
# 
Thanks very much, Dennis.  See below for something I don't understand.
On 10/21/2011 12:15 PM, Dennis Murphy wrote:
This seems to work, but I don't understand *why*, since the weight 
variable is
PearsonLee$frequency, not PearsonLee$weight.

 > PearsonLee$weight
NULL

I get an error if I try to use PearsonLee$frequency as the weights= 
variable.

 > ggplot(PearsonLee, aes(x = parent, y = child)) +
+    geom_point(size = 1.5, position = position_jitter(width = 0.2)) +
+    geom_smooth(method = lm, aes(weights = PearsonLee$frequency),
+                colour = 'green', se = FALSE, size = 1.5) +
+    geom_smooth(aes(weights = PearsonLee$frequency),
+                colour = 'red', se = FALSE, size = 1.5) +
+    facet_grid(chl ~ par)
Error in eval(expr, envir, enclos) : object 'weight' not found

In the form below, it makes sense to me and does work, using 
weight=frequency in the initial aes(),
and weight= in geom_smooth:

ggplot(PearsonLee, aes(x = parent, y = child, weight=frequency)) +
    geom_point(size = 1.5, position = position_jitter(width = 0.2)) +
    geom_smooth(method = lm, aes(weight = PearsonLee$frequency),
                colour = 'green', se = FALSE, size = 1.5) +
    geom_smooth(aes(weight = PearsonLee$frequency),
                colour = 'red', se = FALSE, size = 1.5) +
    facet_grid(chl ~ par)

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Dennis Murphy
    

    

      
      #
    

  


  

      
Hi Michael:

The necessary argument to geom_smooth() is weight, not weights (my
fault, sorry), so try this instead:

ggplot(PearsonLee, aes(x = parent, y = child)) +
   geom_point(size = 1.5, position = position_jitter(width = 0.2)) +
   geom_smooth(method = lm, aes(weight = frequency,
               colour = 'Linear'), se = FALSE, size = 1.5) +
   geom_smooth(aes(weight = frequency,
               colour = 'Loess'), se = FALSE, size = 1.5) +
   facet_grid(chl ~ par) +
   scale_colour_manual(breaks = c('Linear', 'Loess'),
                       values = c('green', 'red')) +
   opts(legend.position = c(0.14, 0.885),
        legend.background = theme_rect(fill = 'white'))

Dennis

        
On Fri, Oct 21, 2011 at 11:57 AM, Michael Friendly <friendly at yorku.ca> wrote: