Wednesday, August 3, 2011

Are students' teaching evaluations influenced by instructors' looks?

Are students' teaching evaluations influenced by instructors' looks? ggplot2 may help find the answer.

The recent release of RcmdrPlugin.KMggplot2 has made ggplot2 available to those who prefer GUI to the command line interface. With the new plugin for Rcmdr, one can simply point and click and within seconds can have an amazing publish-quality graph.

I test drove the plugin and liked it very much accept the glaring omission: it does not provide for faceting for scatter plots, which is a key strength of ggplot2. Let me illustrate this with an example.

I use the Teaching Ratings data from the AER package. The data set was used in a study to determine if students were influenced by the instructor's good looks or otherwise while evaluating the instructor's teaching performance.

Using Rcmdr and the ggplot2 plug-in I quickly draw the graph, which is presented to the right. The blue triangles represent women and the red dots represent male instructors. The Y-axis is a normalized measure of instructor's looks and teaching evaluations are presented on the x-axis.

It appears from the scatter plot that  good looks are correlated with higher teaching evaluations.

Now let's add two more dimensions to account for the following. First, is there a difference in teaching evaluations for tenured instructors versus the non-tenured instructors. And second, do students evaluate instructors belonging to visible minorities groups any different from the rest.

Using faceting in ggplot2, I draw the new graph that is presented to the right. Notice that the graph has four quadrants capturing the 2 x 2 possible combination for tenured (yes/no) and minority group (yes/no).

Let's focus on the top right quadrant, which shows that non-tenured, visible minority faculty members happen to be exclusively males receiving low beauty evaluations, but higher teaching evaluations.

The bottom right quadrant however shows that most tenured visible minority instructors are females who receive a higher teaching and beauty evaluation.

The second graph was produced by tweaking the command generated by the plug-in to illustrate that with faceting one can quickly plot multi-dimensional graphs. I hope that the next version of the plug-in  incorporates faceting for scatter plots.

Below is the code used to generate the second graph.

data(TeachingRatings, package="AER")

.df <- data.frame(x = tr$eval,y = tr$beauty,z = tr$gender, z1=tr$tenure, z2=tr$minority)

.plot1 <- ggplot(data = .df, aes(x = x, y = y, colour = z, shape = z)) + geom_point() +
  scale_colour_brewer(palette="Set1") +facet_wrap(z1~z2) + xlab("eval") + ylab("beauty") + labs(colour = "gender", shape = "gender") + kmg2_theme_gray(10, "sans") + opts(legend.position =
rm(.df, .plot1)


  1. Nice plots. I also like this data set. For a model-based analysis using recursive partitioning, see the following presentation:

  2. This reminds me of the research reported in the book "Dress for Success." I think his first research project was to have the male teachers wear a tie in some semesters and not in others. Rather than teacher evaluations he studied resulting grades and number of student behavioral problems. Both got better in the semesters where the teacher wore a tie. That book reports the results of many such studies showing the impact of clothes on perceived intelligence, trustworthiness, etc, though I suspect they're quite dated.

    Bob Muenchen

  3. Hello, I'm the maintainer of the RcmdrPlugin.KMggplot2 package. Thank you for your comment. I would like to implement your suggestion in the next release.