Many college courses conclude by giving students the opportunity to evaluate the course and the instructor anonymously. However, the use of these student evaluations as an indicator of course quality and teaching effectiveness is often criticized because these measures may reflect the influence of non-teaching related characteristics, such as the physical appearance of the instructor. The article titled, “Beauty in the classroom: instructors’ pulchritude and putative pedagogical productivity” by Hamermesh and Parker found that instructors who are viewed to be better looking receive higher instructional ratings.
Here, you will analyze the data from this study in order to learn what goes into a positive professor evaluation.
In this lab, you will explore and visualize the data using the tidyverse suite of packages. The data can be found in the companion package for OpenIntro resources, openintro.
Let’s load the packages.
This is the first time we’re using the GGally
package.
You will be using the ggpairs
function from this package
later in the lab.
The data were gathered from end of semester student evaluations for a
large sample of professors from the University of Texas at Austin. In
addition, six students rated the professors’ physical appearance. The
result is a data frame where each row contains a different course and
columns represent variables about the courses and professors. It’s
called evals
.
## Rows: 463
## Columns: 23
## $ course_id <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 1…
## $ prof_id <int> 1, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 5, 5,…
## $ score <dbl> 4.7, 4.1, 3.9, 4.8, 4.6, 4.3, 2.8, 4.1, 3.4, 4.5, 3.8, 4…
## $ rank <fct> tenure track, tenure track, tenure track, tenure track, …
## $ ethnicity <fct> minority, minority, minority, minority, not minority, no…
## $ gender <fct> female, female, female, female, male, male, male, male, …
## $ language <fct> english, english, english, english, english, english, en…
## $ age <int> 36, 36, 36, 36, 59, 59, 59, 51, 51, 40, 40, 40, 40, 40, …
## $ cls_perc_eval <dbl> 55.81395, 68.80000, 60.80000, 62.60163, 85.00000, 87.500…
## $ cls_did_eval <int> 24, 86, 76, 77, 17, 35, 39, 55, 111, 40, 24, 24, 17, 14,…
## $ cls_students <int> 43, 125, 125, 123, 20, 40, 44, 55, 195, 46, 27, 25, 20, …
## $ cls_level <fct> upper, upper, upper, upper, upper, upper, upper, upper, …
## $ cls_profs <fct> single, single, single, single, multiple, multiple, mult…
## $ cls_credits <fct> multi credit, multi credit, multi credit, multi credit, …
## $ bty_f1lower <int> 5, 5, 5, 5, 4, 4, 4, 5, 5, 2, 2, 2, 2, 2, 2, 2, 2, 7, 7,…
## $ bty_f1upper <int> 7, 7, 7, 7, 4, 4, 4, 2, 2, 5, 5, 5, 5, 5, 5, 5, 5, 9, 9,…
## $ bty_f2upper <int> 6, 6, 6, 6, 2, 2, 2, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 9, 9,…
## $ bty_m1lower <int> 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 7, 7,…
## $ bty_m1upper <int> 4, 4, 4, 4, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 6, 6,…
## $ bty_m2upper <int> 6, 6, 6, 6, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 2, 6, 6,…
## $ bty_avg <dbl> 5.000, 5.000, 5.000, 5.000, 3.000, 3.000, 3.000, 3.333, …
## $ pic_outfit <fct> not formal, not formal, not formal, not formal, not form…
## $ pic_color <fct> color, color, color, color, color, color, color, color, …
We have observations on 21 different variables, some categorical and some numerical. The meaning of each variable can be found by bringing up the help file:
This is strictly an observational study based upon obtained data. There is nothing experimental about the study, as there is no intervention being implemented to qualify. Beauty, of course, is in the eye of the beholder and not necessarily an objective measure that is uniformly assessed by the students. Also, one would have to be sure that there is a direct pathway, not confounded, mediated, or moderated, between the perception of beauty by the students and a noted difference in the course evaluations. You would have to control for all possible confounding, and based upon this current study, I would be concerned with that presumption. I would suggest changing it the research question to be, given the possible co-variables, and all things remaining constant, is there correlation between student perceived beauty of the professor and their course evaluation scores.
score
. Is the distribution
skewed? What does that tell you about how students rate courses? Is this
what you expected to see? Why, or why not?
By plotting a histogram we note that there seems to be a
left-hand skewing of the distribution. Immediately, that tells us that
in general the students rate the professors higher overall. I am not
surprised because in general students who do poorly in a particular
course would tend to rate the professor lower and, since most students
tend to pass courses their professor evaluations would also tend to be
higher. For this particular study, we would have to determine whether
“beauty” has a particular affect contributing to the distribution. We
can only speak about the general population of professors and not any
particular professor.
score
, select two other variables and
describe their relationship with each other using an appropriate
visualization.ggplot(evals, aes(age, bty_avg, color=gender)) + geom_point() + geom_smooth(method = "lm") + facet_wrap(~gender)
In these visualizations, we are exploring the potential relationship between age and the average beauty score of the professor. We can see by the first visualization that there seems to be a inverse relationship with age and average beauty score. When we control for gender, we see that the the curve may be slightly steeper for males vs. females, but there are also more older males over 65 than females, which could be biasing the curve downward a bit.
The fundamental phenomenon suggested by the study is that better looking teachers are evaluated more favorably. Let’s create a scatterplot to see if this appears to be the case:
Before you draw conclusions about the trend, compare the number of observations in the data frame with the approximate number of points on the scatterplot. Is anything awry?
geom_jitter
as your layer. What was misleading about the initial scatterplot?The discreteness of the data points an indication that there may be an increasing linear trend between evaluation score and the professors perceived beauty score. By using geom_jitter, a small random variation is added to the location of each data point removing a potential optical illusion based upon the discreteness. It now looks less organized.
m_bty
to
predict average professor score by average beauty rating. Write out the
equation for the linear model and interpret the slope. Is average beauty
score a statistically significant predictor? Does it appear to be a
practically significant predictor?##
## Call:
## lm(formula = score ~ bty_avg, data = evals)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.9246 -0.3690 0.1420 0.3977 0.9309
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.88034 0.07614 50.96 < 2e-16 ***
## bty_avg 0.06664 0.01629 4.09 5.08e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5348 on 461 degrees of freedom
## Multiple R-squared: 0.03502, Adjusted R-squared: 0.03293
## F-statistic: 16.73 on 1 and 461 DF, p-value: 5.083e-05
Though the result is statistically significant (p < 0.00005), the effect size is relatively small. For every 1 point increase in average beauty, the evaluation score goes up by 0.067 points. Also, this model only accounts for approximately 3% of the variance noted in the response variable (score). Not a very good predictor.
Add the line of the bet fit model to your plot using the following:
The blue line is the model. The shaded gray area around the line
tells you about the variability you might expect in your predictions. To
turn that off, use se = FALSE
.
ggplot(data = evals, aes(x = bty_avg, y = score)) +
geom_jitter() +
geom_smooth(method = "lm", se = FALSE)
ggplot(data = m_bty, aes(x = .fitted, y = .resid)) +
geom_point() +
geom_hline(yintercept = 0, linetype = "dashed", color="red") +
xlab("Fitted values") +
ylab("Residuals")
Looking at the various plots note some potential issues, but for
the most part they appear to indicate normality. It is important to note
the issues. The residuals plot indicates a larger number of residuals
below the base line that could indicate a heavier weighting. The highest
peak of the residuals plot is not centered at 0, indicating a bias in
the model. This could be due to the outliers noticed in the plot. The
histogram of the residuals indicate a leftward skewing which support and
coincides with the residual plot. The QQ plot shows a slight concavity
downward, but it seems fairly normal.
The data set contains several variables on the beauty score of the professor: individual ratings from each of the six students who were asked to score the physical appearance of the professors and the average of these six scores. Let’s take a look at the relationship between one of these scores and the average beauty score.
## # A tibble: 1 × 1
## `cor(bty_avg, bty_f1lower)`
## <dbl>
## 1 0.844
As expected, the relationship is quite strong—after all, the average score is calculated using the individual scores. You can actually look at the relationships between all beauty variables (columns 13 through 19) using the following command:
These variables are collinear (correlated), and adding more than one of these variables to the model would not add much value to the model. In this application and with these highly-correlated predictors, it is reasonable to use the average beauty score as the single representative of these variables.
In order to see if beauty is still a significant predictor of professor score after you’ve accounted for the professor’s gender, you can add the gender term into the model.
##
## Call:
## lm(formula = score ~ bty_avg + gender, data = evals)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8305 -0.3625 0.1055 0.4213 0.9314
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.74734 0.08466 44.266 < 2e-16 ***
## bty_avg 0.07416 0.01625 4.563 6.48e-06 ***
## gendermale 0.17239 0.05022 3.433 0.000652 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5287 on 460 degrees of freedom
## Multiple R-squared: 0.05912, Adjusted R-squared: 0.05503
## F-statistic: 14.45 on 2 and 460 DF, p-value: 8.177e-07
ggplot(data = m_bty_gen, aes(x = .fitted, y = .resid)) +
geom_point() +
geom_hline(yintercept = 0, linetype = "dashed", color="red") +
xlab("Fitted values") +
ylab("Residuals")
Like the previous single variable model, we see the curves appear to fairly normal with a left hand skew and a concave downward QQ plot. We also see a farther peak to the right off 0, indicating a bias and potential outlier impact on the model. However, it appears to be fairly normal and does not violate the normality required for OLS.
bty_avg
still a significant predictor of
score
? Has the addition of gender
to the model
changed the parameter estimate for bty_avg
?Yes, bty_avg is still a predictor of score. Yes, the parameter estimate increased in magnitude, but is still relatively small. Also, the amount of variance in the response variable due to the independent variables (R squared) improved from about 3% to 5%, but is still pretty weak.
Note that the estimate for gender
is now called
gendermale
. You’ll see this name change whenever you
introduce a categorical variable. The reason is that R recodes
gender
from having the values of male
and
female
to being an indicator variable called
gendermale
that takes a value of \(0\) for female professors and a value of
\(1\) for male professors. (Such
variables are often referred to as “dummy” variables.)
As a result, for female professors, the parameter estimate is multiplied by zero, leaving the intercept and slope form familiar from simple regression.
\[ \begin{aligned} \widehat{score} &= \hat{\beta}_0 + \hat{\beta}_1 \times bty\_avg + \hat{\beta}_2 \times (0) \\ &= \hat{\beta}_0 + \hat{\beta}_1 \times bty\_avg\end{aligned} \]
ggplot(data = evals, aes(x = bty_avg, y = score, color = pic_color)) +
geom_smooth(method = "lm", formula = y ~ x, se = FALSE)
\[ \begin{aligned} \widehat{score} &= \hat{\beta}_0 + \hat{\beta}_1 \times bty\_avg + \hat{\beta}_2 \times pic\_color \\ &= \hat{\beta}_0 + \hat{\beta}_1 \times bty\_avg + \hat{\beta}_2 \times (1)\end{aligned} \]
Equation provided above. So, two people receiving an average beauty rating of 4, the person with the black and white picture would have the higher evaluation score. The person with the color picture would have a lower evaluation score. In this case, since black&white comes alphabetically before color, that is given the value 0 and color is given the value 1. But since this pushes the curve down, we can deduce that the parameter estimate (beta2) has a negative impact on the evaluation score.
The decision to call the indicator variable gendermale
instead of genderfemale
has no deeper meaning. R simply
codes the category that comes first alphabetically as a \(0\). (You can change the reference level of
a categorical variable, which is the level that is coded as a 0, using
therelevel()
function. Use ?relevel
to learn
more.)
m_bty_rank
with
gender
removed and rank
added in. How does R
appear to handle categorical variables that have more than two levels?
Note that the rank variable has three levels: teaching
,
tenure track
, tenured
.##
## Call:
## lm(formula = score ~ bty_avg + rank, data = evals)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8713 -0.3642 0.1489 0.4103 0.9525
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.98155 0.09078 43.860 < 2e-16 ***
## bty_avg 0.06783 0.01655 4.098 4.92e-05 ***
## ranktenure track -0.16070 0.07395 -2.173 0.0303 *
## ranktenured -0.12623 0.06266 -2.014 0.0445 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5328 on 459 degrees of freedom
## Multiple R-squared: 0.04652, Adjusted R-squared: 0.04029
## F-statistic: 7.465 on 3 and 459 DF, p-value: 6.88e-05
R handles the new categorical variable with three levels, similar to that of two. The model is a simple univariate equation when the variable is “teaching” and then the beta takes on different value when the variable is either “tenure track” or “tenured”.
The interpretation of the coefficients in multiple regression is
slightly different from that of simple regression. The estimate for
bty_avg
reflects how much higher a group of professors is
expected to score if they have a beauty rating that is one point higher
while holding all other variables constant. In this case, that
translates into considering only professors of the same rank with
bty_avg
scores that are one point apart.
We will start with a full model that predicts professor score based on rank, gender, ethnicity, language of the university where they got their degree, age, proportion of students that filled out evaluations, class size, course level, number of professors, number of credits, average beauty rating, outfit, and picture color.
Without peeking, I would think a few variables might qualify. I suspect cls_level and cls_students might be highly correlated as upper division courses tend to have fewer students than lower level courses. I also suspect the cls_profs might qualify as it does not seem rational that the sections of a course being taught by a number professors would matter, as one would think that evaluations would be independent as a student would not be sitting in two sections of the same course to make comparative evaluations.
Let’s run the model…
m_full <- lm(score ~ rank + gender + ethnicity + language + age + cls_perc_eval
+ cls_students + cls_level + cls_profs + cls_credits + bty_avg
+ pic_outfit + pic_color, data = evals)
summary(m_full)
##
## Call:
## lm(formula = score ~ rank + gender + ethnicity + language + age +
## cls_perc_eval + cls_students + cls_level + cls_profs + cls_credits +
## bty_avg + pic_outfit + pic_color, data = evals)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.77397 -0.32432 0.09067 0.35183 0.95036
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.0952141 0.2905277 14.096 < 2e-16 ***
## ranktenure track -0.1475932 0.0820671 -1.798 0.07278 .
## ranktenured -0.0973378 0.0663296 -1.467 0.14295
## gendermale 0.2109481 0.0518230 4.071 5.54e-05 ***
## ethnicitynot minority 0.1234929 0.0786273 1.571 0.11698
## languagenon-english -0.2298112 0.1113754 -2.063 0.03965 *
## age -0.0090072 0.0031359 -2.872 0.00427 **
## cls_perc_eval 0.0053272 0.0015393 3.461 0.00059 ***
## cls_students 0.0004546 0.0003774 1.205 0.22896
## cls_levelupper 0.0605140 0.0575617 1.051 0.29369
## cls_profssingle -0.0146619 0.0519885 -0.282 0.77806
## cls_creditsone credit 0.5020432 0.1159388 4.330 1.84e-05 ***
## bty_avg 0.0400333 0.0175064 2.287 0.02267 *
## pic_outfitnot formal -0.1126817 0.0738800 -1.525 0.12792
## pic_colorcolor -0.2172630 0.0715021 -3.039 0.00252 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.498 on 448 degrees of freedom
## Multiple R-squared: 0.1871, Adjusted R-squared: 0.1617
## F-statistic: 7.366 on 14 and 448 DF, p-value: 6.552e-14
While the cls_level and cls_students are not statistically significant, they are very close to each others p_values. However, they are not the highest. It does appear the cls_profs does have the highest p_value of 0.77806.
With everything else remaining constant, a professor who is “not minority” could expect their evaluation score to increase by a value of 0.1234929; however, that result is not statistically significant (p=0.11698). Therefore, it could indicate that the effect on the evaluation score could be “0”.
m_full_minus_1 <- lm(score ~ rank + gender + ethnicity + language + age + cls_perc_eval
+ cls_students + cls_level + cls_credits + bty_avg
+ pic_outfit + pic_color, data = evals)
summary(m_full_minus_1)
##
## Call:
## lm(formula = score ~ rank + gender + ethnicity + language + age +
## cls_perc_eval + cls_students + cls_level + cls_credits +
## bty_avg + pic_outfit + pic_color, data = evals)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.7836 -0.3257 0.0859 0.3513 0.9551
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.0872523 0.2888562 14.150 < 2e-16 ***
## ranktenure track -0.1476746 0.0819824 -1.801 0.072327 .
## ranktenured -0.0973829 0.0662614 -1.470 0.142349
## gendermale 0.2101231 0.0516873 4.065 5.66e-05 ***
## ethnicitynot minority 0.1274458 0.0772887 1.649 0.099856 .
## languagenon-english -0.2282894 0.1111305 -2.054 0.040530 *
## age -0.0089992 0.0031326 -2.873 0.004262 **
## cls_perc_eval 0.0052888 0.0015317 3.453 0.000607 ***
## cls_students 0.0004687 0.0003737 1.254 0.210384
## cls_levelupper 0.0606374 0.0575010 1.055 0.292200
## cls_creditsone credit 0.5061196 0.1149163 4.404 1.33e-05 ***
## bty_avg 0.0398629 0.0174780 2.281 0.023032 *
## pic_outfitnot formal -0.1083227 0.0721711 -1.501 0.134080
## pic_colorcolor -0.2190527 0.0711469 -3.079 0.002205 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4974 on 449 degrees of freedom
## Multiple R-squared: 0.187, Adjusted R-squared: 0.1634
## F-statistic: 7.943 on 13 and 449 DF, p-value: 2.336e-14
While some p_values increase or decreased, the overall remaining values did not change enough to alter their original statistical significance in the full model. However, the adjusted-R square did increase slightly.
m_full_minus_all <- lm(score ~ gender + ethnicity + age + cls_perc_eval
+ cls_credits + bty_avg
+ pic_color, data = evals)
summary(m_full_minus_all)
##
## Call:
## lm(formula = score ~ gender + ethnicity + age + cls_perc_eval +
## cls_credits + bty_avg + pic_color, data = evals)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.85434 -0.33568 0.09247 0.38288 0.93903
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.690771 0.229181 16.104 < 2e-16 ***
## gendermale 0.201574 0.050220 4.014 6.99e-05 ***
## ethnicitynot minority 0.216955 0.071348 3.041 0.00250 **
## age -0.006034 0.002621 -2.302 0.02176 *
## cls_perc_eval 0.004719 0.001439 3.278 0.00113 **
## cls_creditsone credit 0.527806 0.103839 5.083 5.44e-07 ***
## bty_avg 0.052431 0.016975 3.089 0.00213 **
## pic_colorcolor -0.170149 0.066780 -2.548 0.01116 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5008 on 455 degrees of freedom
## Multiple R-squared: 0.1649, Adjusted R-squared: 0.1521
## F-statistic: 12.84 on 7 and 455 DF, p-value: 4.344e-15
Score = B0 + B1 x gender + B2 x ethnicity + B3 x age + B4 x cls_per_eval + B5 x cls_credit + B6 x bty_avg + B7 x pic_color
ggplot(data = m_full_minus_all, aes(x = .fitted, y = .resid)) +
geom_point() +
geom_hline(yintercept = 0, linetype = "dashed", color="red") +
xlab("Fitted values") +
ylab("Residuals")
ggplot(data = m_full_minus_all, aes(x = .resid)) +
geom_histogram(binwidth = 0.2) +
xlab("Residuals")
While we were able to fit the model and get a result, I am
becoming more concerned about the results of the diagnostic plots. The
residual plot appears to be taking on a cone shaped appearance,
indicating that residuals are becoming closer to 0 as the fitted values
increase in magnitude and there are many higher magnitude negative
outliers appearing. The histogram of the residuals indicates a left hand
skew, with a peak farther to the right away from 0. Also the QQ plot
seems to be taking on a greater concavity downward than our previous
model runs. We may have a less than optimal model.
Absolutely. That would mean that each row (observation) may be highly correlated with one or more other rows, and be influencing the results. This may explain the disparities in noted previously with the residuals. Logically, you would have to calculate a mean for each of the variables for a particular professor and then use that in the modelling.
Based upon the model and variable significance, positive characteristics that would increase the score would be male, not minority, greater percent of class completing evaluation, higher credit course, and higher average beauty rating. The attributes that would have the least negative impact on the evaluation score would be a younger professor who had a black and white picture. However, it should be noted that two variables, though significant, had very small impacts on the overall evaluation score in this model-age and percent of class completing the evaluation.
Absolutely not. Besides the fact the model is not optimal, this was a single population sampled under a single university academic structure. We do not know anything about the population of students being asked to complete the evaluations, nor do we know if the academic faculty is consistent with other universities. The best we can ascertain from this modeling and analysis is that it applies only to this specific population and is not generalizable to a larger population without additional information and data sampling.