Ensemble Weightings and Metric Importance

Coh-Metrix Model 3per

This model used Coh-Metrix scores from 15 min persuasive writing samples from 202 students in Grades 2-5 to predict holistic writing quality on the samples (theta scores 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).

Algorithm Weightings in Ensemble

Abbreviations:

• overall = ensemble model
• pls = partial least squares regression
• gbm = stochastic gradient boosted trees
• mars = bagged multivariate adaptive regression splines
• cube = cubist regression

The table below presents the linear weightings of each algorithm for the ensemble model.

Intercept	pls	mars	gbm	cube
-0.0381	0.0558	0.4924	0.4425	0.0259

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	mars	gbm	cube
DESWC	32.09	4.68	34.34	33.8	19.05
WRDHYPn	10.41	2.03	17.13	4.3	5.17
LDVOCD	9.59	3.43	0	21.44	2.53
DESWLlt	8.27	1.29	13.16	3.77	7.4
LSAGN	6.13	2.92	8.63	3.88	4.05
WRDNOUN	4.46	1.39	7.36	1.66	3.95
WRDADV	3.05	1.18	5.53	0.52	3.24
WRDFRQa	2.13	0.25	3.86	0.55	0.2
SMCAUSwn	2.11	1.08	3.7	0.54	1.01
CNCAdd	1.76	0.66	3.37	0.19	0.2
WRDADJ	1.67	0.59	2.93	0.26	4.15
LDTTRa	1.47	4.05	0	2.58	4.76
DESSC	1.38	3.27	0	2.6	2.74
LDTTRc	1.17	3.17	0	2.21	1.32
DESWLltd	0.63	1.31	0	1.2	1.42
WRDPRO	0.57	1.57	0	1.1	0.3
PCDCz	0.5	1.55	0	0.94	0.3
DESSLd	0.47	1.61	0	0.84	0.61
CRFCWO1d	0.47	2.36	0	0.74	0.81
DRNEG	0.46	1.75	0	0.72	1.82
WRDPOLc	0.46	0.78	0	0.89	1.22
SYNNP	0.45	0.82	0	0.92	0
CNCCaus	0.44	0.24	0	0.98	0
WRDPRP3p	0.4	0.59	0	0.82	0.3
WRDHYPv	0.39	0.94	0	0.76	0.2
LDMTLD	0.35	0.38	0	0.67	1.72
WRDAOAc	0.34	1.9	0	0.52	0.3
CNCPos	0.32	0.48	0	0.66	0
WRDMEAc	0.32	1.11	0	0.56	0.61
CRFANPa	0.31	1.37	0	0.52	0.2
DRVP	0.31	0.47	0	0.58	1.22
WRDFRQc	0.31	0.22	0	0.65	0.71
SMCAUSlsa	0.26	0.65	0	0.5	0
LSASS1d	0.26	2.04	0	0.34	0
CRFAO1	0.25	2.03	0	0.3	0.2
WRDHYPnv	0.25	0.33	0	0.48	0.71
CNCTempx	0.23	0.5	0	0.32	2.23
SYNSTRUTa	0.22	0.75	0	0.39	0.41
SYNMEDpos	0.21	2.04	0	0.14	1.42
WRDPRP1s	0.2	0.81	0	0.26	1.62
PCVERBz	0.2	1.75	0	0.15	1.62
CRFCWO1	0.19	1.81	0	0.18	0.41
RDL2	0.19	1.54	0	0.22	0.51
PCNARz	0.18	2.01	0	0.05	1.72
LSAGNd	0.18	2.08	0	0.1	0.91
RDFKGL	0.17	0.52	0	0.19	2.13
PCSYNz	0.17	0.67	0	0.19	2.13
LSASSpd	0.15	1.95	0	0.08	0.2
SMCAUSv	0.15	0.37	0	0.26	0.61
SYNLE	0.14	0.44	0	0.26	0
RDFRE	0.14	0.45	0	0.14	2.13
DRNP	0.14	0.41	0	0.28	0
PCDCp	0.14	1.89	0	0	1.62
SMCAUSr	0.13	1.29	0	0.13	0
LSASS1	0.13	1.84	0	0.04	0.41
CRFCWOad	0.13	1.93	0	0.04	0.3
WRDPRP2	0.12	1.32	0	0.1	0
WRDFAMc	0.12	0.32	0	0.23	0.1
PCREFz	0.12	1.13	0	0.06	1.42
DESWLsy	0.11	0.61	0	0.13	0.51
CNCLogic	0.11	0.55	0	0.17	0.1
CRFNO1	0.11	1.61	0	0.04	0.2
DRGERUND	0.1	0.38	0	0.16	0.41
DRPP	0.1	0.34	0	0.19	0
PCTEMPp	0.1	1.16	0	0.07	0.2
SMINTEr	0.1	1.5	0	0.03	0.2
PCCNCz	0.1	1.31	0	0.05	0.41
DESWLsyd	0.1	0.76	0	0.13	0.3
CNCNeg	0.09	0.55	0	0.13	0.3
WRDVERB	0.08	0.23	0	0.15	0
SMCAUSvp	0.08	0.24	0	0.14	0.2
PCCONNz	0.08	0.49	0	0.11	0.3
PCCONNp	0.07	1.11	0	0.01	0
DRAP	0.07	0.12	0	0.14	0
WRDCNCc	0.07	0.03	0	0.14	0.2
WRDFRQmc	0.07	1.06	0	0.03	0
PCVERBp	0.07	1.19	0	0	0.3
PCREFp	0.06	0.92	0	0	0.3
PCCNCp	0.05	0.83	0	0	0
WRDPRP1p	0.05	0.16	0	0.09	0
WRDIMGc	0.05	0	0	0.09	0.51
SMINTEp	0.05	0.71	0	0.03	0
DRINF	0.05	0.3	0	0.06	0.3
CNCADC	0.05	0.44	0	0.05	0.2
CNCTemp	0.04	0.58	0	0.02	0
PCSYNp	0.04	0.42	0	0.01	0.71
DRPVAL	0.01	0.07	0	0.01	0