CohMetrix Model 2a Variable Importance - shmercer/writeAlizer GitHub Wiki
Ensemble Weightings and Metric Importance
Coh-Metrix Model 2a
This model used Coh-Metrix 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 Coh-Metrix 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
- gbm = stochastic gradient boosted trees
- 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 | gbm | svm | cube |
---|---|---|---|---|---|---|
-11.0081 | 0.1741 | 0.0413 | 0.1875 | 0.2353 | 0.206 | 0.2108 |
Metric Importance in Each Algorithm and Ensemble
Each column sums to 100 (so values can be interpreted as % contribution to the model).
Detailed information on Coh-Metrix abbreviations and indices is available here.
Metric | overall | pls | rf | mars | gbm | svm | cube |
---|---|---|---|---|---|---|---|
DESWC | 29.91 | 5.86 | 14.51 | 55.78 | 44.87 | 5.96 | 36.49 |
DESWLlt | 8.67 | 2.98 | 2.69 | 18.79 | 3.07 | 2.39 | 17.89 |
LDMTLD | 7.35 | 4.01 | 5.57 | 0 | 9.55 | 3.9 | 17.89 |
WRDHYPn | 7.16 | 2.9 | 2.68 | 9.55 | 3.85 | 2.29 | 17.89 |
LDTTRa | 2.84 | 2.01 | 0.97 | 11.14 | 0.26 | 1.12 | 1.05 |
CNCPos | 1.36 | 0.88 | 1.11 | 4.75 | 0.05 | 1.28 | 0.35 |
DESWLsy | 1.3 | 2.09 | 1.9 | 0 | 1.66 | 1.54 | 1.05 |
CNCTempx | 1.11 | 0.89 | 1.03 | 0 | 1.63 | 2.48 | 0.35 |
CNCLogic | 1.09 | 1.42 | 1.39 | 0 | 1.81 | 1.7 | 0.35 |
PCDCp | 1.08 | 2 | 2.56 | 0 | 1.71 | 1.36 | 0 |
DESPL | 1.04 | 3.22 | 2.18 | 0 | 0.04 | 2.15 | 0 |
DESWLltd | 1.02 | 2.3 | 1.13 | 0 | 1.26 | 1.6 | 0 |
WRDFRQa | 0.96 | 1.86 | 0.58 | 0 | 0.17 | 1.97 | 1.05 |
DESWLsyd | 0.92 | 1.97 | 1.87 | 0 | 1.21 | 1.29 | 0 |
DESSLd | 0.9 | 1.67 | 2 | 0 | 1.27 | 1.36 | 0 |
CNCTemp | 0.89 | 0.93 | 1.39 | 0 | 1.08 | 1.89 | 0.35 |
LSAGN | 0.87 | 2.55 | 1.18 | 0 | 0.22 | 1.8 | 0 |
LSASSpd | 0.86 | 1.95 | 0.86 | 0 | 0.36 | 1.79 | 0.35 |
CNCADC | 0.85 | 1.12 | 1.41 | 0 | 0.67 | 2.37 | 0 |
DRPP | 0.84 | 2.18 | 2.11 | 0 | 0.57 | 1.4 | 0 |
PCCONNz | 0.82 | 1.09 | 0.4 | 0 | 1.61 | 1.36 | 0 |
WRDPRO | 0.82 | 1.77 | 1.44 | 0 | 0.73 | 1.61 | 0 |
SYNSTRUTa | 0.8 | 0.85 | 3.07 | 0 | 0.58 | 1.4 | 0.7 |
CRFCWO1d | 0.79 | 1.67 | 1.39 | 0 | 0.43 | 1.88 | 0 |
LSASS1d | 0.78 | 1.55 | 0.48 | 0 | 0.78 | 1.72 | 0 |
SMCAUSwn | 0.76 | 1.4 | 1.04 | 0 | 0.31 | 2.16 | 0 |
SYNMEDpos | 0.75 | 1.75 | 0.6 | 0 | 0.77 | 1.36 | 0 |
SMINTEp | 0.73 | 0.75 | 0.85 | 0 | 0.79 | 1.66 | 0.35 |
LDTTRc | 0.73 | 1.82 | 0.7 | 0 | 0.72 | 1.23 | 0 |
CRFCWOad | 0.73 | 1.51 | 1.12 | 0 | 0.38 | 1.81 | 0 |
WRDVERB | 0.71 | 1.04 | 0.93 | 0 | 0.58 | 0.86 | 1.05 |
WRDFAMc | 0.7 | 1.09 | 0.5 | 0 | 1.32 | 1.08 | 0 |
WRDHYPnv | 0.69 | 1.9 | 0.91 | 0 | 0.13 | 1.22 | 0.35 |
WRDFRQmc | 0.69 | 1.35 | 2.5 | 0 | 1.3 | 0.4 | 0 |
WRDCNCc | 0.69 | 1.07 | 0.44 | 0 | 0.58 | 0.8 | 1.05 |
PCNARz | 0.68 | 1.29 | 1.29 | 0 | 0.58 | 1.1 | 0.35 |
WRDPOLc | 0.67 | 0.85 | 0.77 | 0 | 0.51 | 1.97 | 0 |
RDFRE | 0.66 | 1.29 | 0.91 | 0 | 0.76 | 1.23 | 0 |
CRFNOa | 0.63 | 0.74 | 1.34 | 0 | 0.56 | 1.67 | 0 |
PCVERBz | 0.63 | 1.55 | 0.67 | 0 | 0.28 | 1.09 | 0.35 |
LSAGNd | 0.61 | 1.5 | 0.95 | 0 | 0.24 | 1.38 | 0 |
SMCAUSvp | 0.59 | 0.93 | 0.73 | 0 | 0.36 | 1.67 | 0 |
WRDADV | 0.58 | 1.3 | 0.95 | 0 | 0.13 | 1.55 | 0 |
WRDAOAc | 0.58 | 1.66 | 0.46 | 0 | 0.27 | 1.18 | 0 |
DRNP | 0.57 | 1.73 | 1.5 | 0 | 0.27 | 0.82 | 0 |
WRDHYPv | 0.56 | 0.28 | 1.01 | 0 | 0.83 | 1.47 | 0 |
DRVP | 0.56 | 0.82 | 0.62 | 0 | 0.82 | 0.75 | 0.35 |
PCNARp | 0.54 | 1.76 | 1.29 | 0 | 0 | 1 | 0 |
CNCCaus | 0.51 | 1.11 | 0.58 | 0 | 0.11 | 1.46 | 0 |
CRFCWOa | 0.49 | 0.46 | 0.56 | 0 | 0.37 | 1.57 | 0 |
SMCAUSr | 0.47 | 1.58 | 1.07 | 0 | 0.47 | 0.29 | 0 |
SMCAUSlsa | 0.46 | 0.24 | 0.86 | 0 | 0.63 | 1.26 | 0 |
LSASSp | 0.46 | 0.79 | 1.01 | 0 | 0.15 | 1.31 | 0 |
SMINTEr | 0.45 | 1.83 | 0.9 | 0 | 0.13 | 0.45 | 0 |
DRAP | 0.44 | 0.46 | 1.01 | 0 | 0.41 | 1.17 | 0 |
DRNEG | 0.44 | 0.76 | 0.43 | 0 | 0.09 | 1.41 | 0 |
WRDNOUN | 0.43 | 0.56 | 0.96 | 0 | 0.75 | 0.66 | 0 |
WRDFRQc | 0.43 | 0.54 | 0.67 | 0 | 0.48 | 1.04 | 0 |
WRDPRP3s | 0.43 | 0.76 | 0.84 | 0 | 0.74 | 0.56 | 0 |
CRFANPa | 0.41 | 0.76 | 1.36 | 0 | 0.18 | 0.96 | 0 |
SYNLE | 0.4 | 1.13 | 0.95 | 0 | 0.1 | 0.77 | 0 |
PCTEMPp | 0.4 | 1.05 | 0.22 | 0 | 0.55 | 0.47 | 0 |
WRDADJ | 0.39 | 0.99 | 0.94 | 0 | 0.2 | 0.74 | 0 |
RDL2 | 0.39 | 0.31 | 1.06 | 0 | 0.39 | 1.09 | 0 |
WRDIMGc | 0.37 | 0.14 | 0.74 | 0 | 0.3 | 0.93 | 0.35 |
PCVERBp | 0.36 | 0.99 | 1.39 | 0 | 0 | 0.75 | 0 |
PCCNCz | 0.35 | 1.18 | 0.1 | 0 | 0.32 | 0.39 | 0 |
WRDMEAc | 0.29 | 0.21 | 0.45 | 0 | 0.44 | 0.71 | 0 |
PCREFz | 0.29 | 0.47 | 1.07 | 0 | 0.3 | 0.53 | 0 |
SMCAUSv | 0.25 | 0.39 | 0.77 | 0 | 0.13 | 0.66 | 0 |
SYNNP | 0.22 | 0.03 | 0.64 | 0 | 0.58 | 0.33 | 0 |
PCSYNp | 0.21 | 0.68 | 1.55 | 0 | 0.15 | 0 | 0 |
PCCONNp | 0.17 | 0 | 0.33 | 0 | 0.11 | 0.66 | 0 |
PCCNCp | 0.16 | 0.57 | 0.78 | 0 | 0 | 0.16 | 0 |
PCREFp | 0.15 | 0.06 | 0.74 | 0 | 0 | 0.58 | 0 |
WRDPRP3p | 0.14 | 0.84 | 0 | 0 | 0 | 0.03 | 0 |