Data on 198 student essays, including the assigned grade and various quantitative writing features such as word length, sentence length, comma usage, grammatical errors, and preposition usage. This dataset is useful for exploring multiple regression with continuous predictors and for model building exercises aimed at predicting essay quality from measurable text features.
Format
A tibble with 198 rows and 8 columns:
- essay
Essay identifier. Type: numeric. Range: (1, 198). Sequential identifier for each essay.
- grade
Essay grade assigned by instructor. Type: numeric. Values: 1, 2, 3, 4. Ordinal scale where higher values indicate better grades (e.g., 1 = lowest, 4 = highest). Mean = 2.28.
- wordlength
Average word length (characters per word). Type: numeric. Range: (3.64, 5.30). Mean = 4.36. Longer average words may indicate more sophisticated vocabulary.
- sqrtwords
Square root of the total word count. Type: numeric. Range: (3.32, 16.79). Mean = 9.38. The square root transformation is used to reduce skewness of the raw word count.
- commas
Comma usage rate (commas per 100 words). Type: numeric. Range: (0.00, 14.70). Mean = 2.48. Higher values indicate more frequent comma use.
- errors
Error rate (grammatical/spelling errors per 100 words). Type: numeric. Range: (0.00, 9.85). Mean = 1.99. Higher values indicate more errors.
- prepos
Preposition usage rate (prepositions per 100 words). Type: numeric. Range: (0.00, 17.80). Mean = 9.78. Moderate preposition use is typical of academic writing.
- sentlength
Average sentence length (words per sentence). Type: numeric. Range: (6.69, 45.45). Mean = 17.41. Longer sentences may indicate more complex writing.
Details
This dataset is used for model building exercises (Chapter 7) and as an example of regression with ordinal outcomes (Chapter 13). Key analyses include: building a multiple regression model to predict essay grade from writing features, forward selection and backward elimination to identify the most important predictors, and ordinal regression treating grade as an ordered response. The dataset illustrates the challenge of predicting a subjective outcome (essay grade) from objective text features.
Examples
data(grades)
head(grades)
#> # A tibble: 6 × 8
#> essay grade wordlength sqrtwords commas errors prepos sentlength
#> <int> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 3 4.76 15.5 5.60 5.55 8 19.5
#> 2 2 3 4.24 9.06 3.60 1.27 9.5 16.4
#> 3 3 1 4.09 16.2 1.10 2.61 14 18.4
#> 4 4 2 4.36 7.55 1.80 1.81 0 14.6
#> 5 5 1 4.31 9.64 2.30 0 10 18.7
#> 6 6 3 4.51 11.9 1.30 0 11.1 20.0
# Multiple regression: predicting grade from writing features
lm(grade ~ wordlength + sqrtwords + errors + sentlength, data = grades)
#>
#> Call:
#> lm(formula = grade ~ wordlength + sqrtwords + errors + sentlength,
#> data = grades)
#>
#> Coefficients:
#> (Intercept) wordlength sqrtwords errors sentlength
#> -3.066849 0.776477 0.225671 -0.124213 0.005582
#>
# Ordinal logistic regression
if (requireNamespace("MASS", quietly = TRUE)) {
MASS::polr(factor(grade) ~ wordlength + errors + sentlength,
data = grades)
}
#> Call:
#> MASS::polr(formula = factor(grade) ~ wordlength + errors + sentlength,
#> data = grades)
#>
#> Coefficients:
#> wordlength errors sentlength
#> 1.21867997 -0.22665852 0.02145606
#>
#> Intercepts:
#> 1|2 2|3 3|4
#> 4.541847 5.620690 6.641347
#>
#> Residual Deviance: 520.9222
#> AIC: 532.9222