Fit a loess line in R on binary data

Question

I'm trying to fit a loess line on a scatterplot of a binary outcome variable with a continuous predictor variable.

Here is the code I am using:

    lw1<-loess(y~x, data=df) 
    plot(y~x, data=df, pch=19, cex=0.1)
    lines(df$x, lw1$fitted, col='blue')

and this is the graph I get: !https://bitbucket.org/heatherjbaldwin/akos_open/src/ec2a78d093e6cdf988434c03c1b7c7df145892ba/loessgraph1.png?at=master

I also tried ordering the x variable:

    j<-order(df$x)
    lines(df$x[j], lw1$fitted, col='blue')

And get this graph: !https://bitbucket.org/heatherjbaldwin/akos_open/src/ec2a78d093e6cdf988434c03c1b7c7df145892ba/loessgraph2%28ordered_x%29.png?at=master

Here is the data: https://bitbucket.org/heatherjbaldwin/akos_open/src/ec2a78d093e6cdf988434c03c1b7c7df145892ba/loesscurvedata.txt?at=master

Any help is much appreciated.

Answer 1

Using ggplot2 and a loess smoother I get this :

ggplot(data=dat,aes(x,y)) + 
     geom_line() + 
     geom_smooth(method='loess')

But I think your a looking for a classifier here.