ReaderBench Model 2a Variable Importance - shmercer/writeAlizer GitHub Wiki
Ensemble Weightings and Metric Importance
ReaderBench Model 2a
This model used ReaderBench scores from 7 min narrative writing samples ("I once had a magic pencil and ...") from 136 students in the fall of Grades 2-5 (Mercer et al., 2019) to predict holistic writing quality on the samples (elo ratings calculated from paired comparisons).
Highly correlated ReaderBench metrics (r > |.90|) were excluded during pre-processing (see section on Scoring Model Development for more details).
Mercer, S. H., Keller-Margulis, M. A., Faith, E. L., Reid, E. K., & Ochs, S. (2019). The potential for automated text evaluation to improve the technical adequacy of written expression curriculum-based measurement. Learning Disability Quarterly, 42, 117-128. https://doi.org/10.1177/0731948718803296
Algorithm Weightings in Ensemble
Abbreviations:
- all = ensemble model
- pls = partial least squares regression
- rf = random forest regression
- mars = bagged multivariate adaptive regression splines
- svm = support vector machines
- cube = cubist regression
The table below presents the linear weightings of each algorithm for the ensemble model.
Intercept | pls | rf | mars | svm | cube |
---|---|---|---|---|---|
-4.338 | 0.2371 | 0.1755 | 0.1780 | 0.2234 | 0.2532 |
Metric Importance in Each Algorithm and Ensemble
Each column sums to 100 (so values can be interpreted as % contribution to the model).
Metric | overall | pls | rf | mars | svm | cube |
---|---|---|---|---|---|---|
WdEnt | 20.53 | 4.67 | 10.12 | 73.84 | 5.16 | 18.67 |
AvgDepsSen_dep | 4.65 | 1.23 | 0.88 | 16.82 | 0.88 | 5.25 |
Content.words | 4.59 | 4.32 | 4.77 | 0 | 4.68 | 7.87 |
Words | 3.72 | 4.44 | 4.67 | 0 | 4.67 | 4.17 |
LxcDiv | 3.1 | 4.08 | 3.29 | 0 | 4.06 | 3.4 |
AvgAOASen_Shock | 2.77 | 1.45 | 1.16 | 9.34 | 1.39 | 1.7 |
TCorefChainDoc | 2.62 | 2.98 | 0.81 | 0 | 2.06 | 5.86 |
AvgChainSpan | 2.59 | 3.27 | 3.83 | 0 | 3.1 | 2.47 |
WdDiffWdStem | 2.46 | 2.73 | 3.07 | 0 | 2.24 | 3.7 |
SynSoph | 2.12 | 1.71 | 0.92 | 0 | 1.82 | 5.09 |
AvgDepsSen_punct | 2.03 | 2.48 | 1.68 | 0 | 1.74 | 3.55 |
TActCorefChainWd | 1.93 | 1.6 | 1.91 | 0 | 1.47 | 4.01 |
WdDiffLemmaStem | 1.66 | 1.52 | 0.72 | 0 | 2.44 | 2.93 |
RdbltyFlesch | 1.55 | 0.77 | 1.22 | 0 | 1.09 | 4.01 |
WdLettStdDev | 1.44 | 2.35 | 1.51 | 0 | 2.13 | 0.93 |
AvgAOESen_InverseAverage | 1.37 | 1.44 | 1.22 | 0 | 1.09 | 2.62 |
Sentences | 1.3 | 2.84 | 1.77 | 0 | 1.82 | 0 |
AvgWdLen | 1.27 | 2.65 | 1.57 | 0 | 2.02 | 0 |
LexChainMaxSp | 1.26 | 2.89 | 1.19 | 0 | 2.02 | 0 |
AvgAOADoc_Shock | 1.26 | 2.36 | 1.89 | 0 | 1.68 | 0.31 |
AvgAOADoc_Kuperman | 1.25 | 0.72 | 1.01 | 0 | 1.3 | 2.78 |
WdSylCnt | 1.15 | 1.57 | 1.83 | 0 | 1.51 | 0.77 |
CharEnt | 1.14 | 2.65 | 0.96 | 0 | 1.85 | 0 |
LexChainAvgSpan | 1.12 | 2.18 | 1.5 | 0 | 1.86 | 0 |
AvgDepsSen_advcl | 1.07 | 0.93 | 0.85 | 0 | 1.35 | 1.85 |
AvgAOASen_Kuperman | 1.04 | 1.23 | 1.48 | 0 | 1.46 | 0.93 |
AvgCorefChain | 1 | 1.86 | 0.85 | 0 | 0.9 | 1.08 |
WdAvgDpthHypernymTree | 1 | 1.14 | 0.87 | 0 | 0.97 | 1.7 |
SenStdDevWd | 0.98 | 1.96 | 1.43 | 0 | 1.49 | 0 |
TCorefChainBigSpan | 0.95 | 2.16 | 1.44 | 0 | 1.13 | 0 |
AvgAOADoc_Bristol | 0.94 | 1.75 | 1.03 | 0 | 1.1 | 0.62 |
LxcSoph | 0.92 | 1.64 | 1.2 | 0 | 0.85 | 0.77 |
AvgAdverbSen | 0.88 | 0.89 | 1.38 | 0 | 1.46 | 0.62 |
RdbltyDaleChall | 0.87 | 1.75 | 1.63 | 0 | 1 | 0 |
AvgSenAdjCoh_LDA | 0.82 | 1.97 | 0.64 | 0 | 1.33 | 0 |
AvgRhythmUnits | 0.82 | 1.12 | 1.13 | 0 | 1.15 | 0.62 |
FrqRhythmId | 0.8 | 1.69 | 1.07 | 0 | 1.18 | 0 |
AvgAOADoc_Bird | 0.78 | 0.95 | 0.3 | 0 | 1.43 | 0.93 |
AvgVoice | 0.78 | 2.01 | 0.76 | 0 | 0.99 | 0 |
AvgAOADoc_Cortese | 0.77 | 0.69 | 1.3 | 0 | 1.57 | 0.31 |
WdPathCntHypernymTree | 0.71 | 1.45 | 0.84 | 0 | 1.17 | 0 |
AvgConnSen_simple_subordinators | 0.7 | 0.51 | 2.49 | 0 | 0.82 | 0 |
AvgAOASen_Bristol | 0.68 | 0.66 | 0.71 | 0 | 1.29 | 0.62 |
AvgRhythmUnitStreesSyll | 0.63 | 0.08 | 0.91 | 0 | 0.81 | 1.23 |
AvgInferenceDistChain | 0.62 | 1.39 | 0.34 | 0 | 1.2 | 0 |
AggPronSen_indefinite | 0.62 | 0.45 | 0.63 | 0 | 1.31 | 0.62 |
AvgAOASen_Bird | 0.6 | 1.13 | 0.37 | 0 | 1.37 | 0 |
AvgDepsSen_compound | 0.6 | 0.72 | 0.5 | 0 | 0.48 | 1.08 |
WdPolysemyCnt | 0.58 | 0 | 1.09 | 0 | 1.9 | 0 |
AvgDepsSen_ccomp | 0.57 | 0.09 | 1.32 | 0 | 0.9 | 0.62 |
AvgAOASen_Cortese | 0.55 | 1.15 | 0.3 | 0 | 1.17 | 0 |
AvgDepsSen_cop | 0.54 | 0.24 | 0.58 | 0 | 0.97 | 0.77 |
AvgPronounSen | 0.54 | 0.12 | 0.93 | 0 | 0.48 | 1.08 |
AvgNmdEntSen | 0.52 | 0.24 | 1.12 | 0 | 1.33 | 0 |
AvgNounSen | 0.52 | 0.24 | 0.15 | 0 | 0.18 | 1.7 |
AvgDepsSen_nmod | 0.48 | 0.7 | 0.69 | 0 | 1 | 0 |
AvgDepsSen_aux | 0.48 | 0.24 | 0.92 | 0 | 1.31 | 0 |
AvgConnSen_addition | 0.48 | 1.1 | 0.6 | 0 | 0.66 | 0 |
AvgDepsSen_dobj | 0.48 | 0.23 | 1.51 | 0 | 0.16 | 0.62 |
AvgAOEDoc_InverseLinearRegressionSlope | 0.44 | 0.4 | 0.8 | 0 | 0.68 | 0.31 |
AvgDepsSen_mark | 0.41 | 0.43 | 0.95 | 0 | 0.73 | 0 |
AvgConnSen_temporal_connectors | 0.41 | 0.32 | 0.64 | 0 | 1.11 | 0 |
AvgDepsSen_det | 0.4 | 0.18 | 0.4 | 0 | 0.72 | 0.62 |
AvgConnSen_semi_coordinators | 0.38 | 0.8 | 0.15 | 0 | 0.16 | 0.62 |
AvgConnSen_order | 0.36 | 0.31 | 1.74 | 0 | 0.03 | 0 |
AggPronSen_third_person | 0.36 | 0.57 | 0.91 | 0 | 0.41 | 0 |
LangRhythmDiameter | 0.35 | 0.57 | 0.79 | 0 | 0.08 | 0.31 |
SenAsson | 0.35 | 0.8 | 0.83 | 0 | 0.16 | 0 |
AvgAOEDoc_IndexAboveThreshold.0.3. | 0.33 | 0.03 | 0.43 | 0 | 0.87 | 0.31 |
AvgDepsSen_amod | 0.29 | 0.33 | 0.98 | 0 | 0.27 | 0 |
AvgAdjectiveSen | 0.28 | 0.1 | 1.28 | 0 | 0.21 | 0 |
AvgConnSen_oppositions | 0.27 | 0.54 | 0.82 | 0 | 0.07 | 0 |
AvgDepsSen_xcomp | 0.24 | 0.01 | 0.13 | 0 | 1.04 | 0 |
AvgAOEDoc_IndexPolynomialFitAboveThreshold.0.3. | 0.21 | 0.12 | 0.1 | 0 | 0.78 | 0 |
LangRhythmId | 0.19 | 0.47 | 0.45 | 0 | 0.05 | 0 |
AvgDepsSen_neg | 0.18 | 0.03 | 1.05 | 0 | 0 | 0 |
AvgDepsSen_mwe | 0.17 | 0.38 | 0.47 | 0 | 0.04 | 0 |
LangRhythmCoeff | 0.16 | 0 | 0.22 | 0 | 0.61 | 0 |
AvgDepsSen_acl | 0.06 | 0.25 | 0 | 0 | 0.02 | 0 |