Ensemble Weightings and Metric Importance

CohMetrix Model 3narr

This model used CohMetrix scores from 15 min narrative 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 CohMetrix 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
• mars = bagged multivariate adaptive regression splines
• enet = elastic net regression
• cube = cubist regression

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

Intercept	pls	mars	gbm	enet	cube
0.0000	0.1419	0.3143	0.0729	0.0816	0.1792

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	gbm	mars	enet	cube
DESWC	24.87	4.56	36.54	23.26	28.45	14.11
WRDHYPn	8.87	1.84	2.7	14.41	7.18	4.52
WRDNOUN	7.12	2.27	2.89	10.61	7.1	4.37
DESSL	7	1.2	0.27	14.41	0	0.77
SYNNP	5.71	1.81	3.49	7.61	8.68	2.38
DESWLlt	5.25	0.86	1.94	7.61	5.68	4.14
LDVOCD	4.17	2.88	3.99	6.34	0	0.69
LDTTRa	2.6	3.58	4.54	0	7.74	3.99
SMCAUSwn	2.18	2	5.72	0	3.96	2.15
SYNLE	2.05	0.5	1.59	3.39	0	0.31
WRDPRP1s	1.71	0.5	0.83	2.89	0	0.84
WRDHYPnv	1.55	0.44	0.28	2.61	0	1.61
PCDCp	1.44	1.37	0.04	2.61	0.04	1
PCREFp	1.41	0.55	0	2.65	0	1
PCNARz	1.32	2.01	3.19	0	0	3.14
CRFANPa	1.3	1.59	3.69	0	0	2.3
LSAGN	1.29	2.26	2.08	0	2.54	2.99
WRDFRQmc	0.94	1.95	0.78	0	2.77	2.68
CNCLogic	0.91	0.73	0.37	1.61	0	0.23
PCDCz	0.91	1.12	3.1	0	0	0.77
SYNMEDpos	0.9	1.78	0.13	0	4.2	2.61
SMCAUSlsa	0.84	0.78	0.76	0	1.34	3.37
PCCONNz	0.75	1.46	1.11	0	2.09	1.46
DRPP	0.65	1.2	0.36	0	2.51	1.76
WRDAOAc	0.65	1.77	1.13	0	1.24	1.23
WRDPRO	0.63	1.35	1.42	0	0	1.53
DESPL	0.61	2.71	0.51	0	1.6	1.46
PCREFz	0.59	1.06	0.56	0	2.75	0.92
WRDMEAc	0.53	0.99	1.17	0	0.82	0.84
LDTTRc	0.52	2.28	0.15	0	0	2.68
DRNP	0.49	0.62	0.24	0	1.25	1.92
DESWLsy	0.49	1.25	0.33	0	0.72	1.99
WRDPOLc	0.44	1.45	0.91	0	0	1.07
SMINTEr	0.4	0.35	0.26	0	2.03	0.84
PCSYNz	0.39	0.91	0.59	0	0	1.46
PCCONNp	0.37	2.07	0.2	0	1.9	0.31
SMINTEp	0.35	0.56	1.24	0	0	0.15
DRPVAL	0.34	1.5	0.33	0	1.65	0.23
LSASS1d	0.33	1.49	0.06	0	0	1.76
PCCNCz	0.32	1.44	0.1	0	0	1.61
CRFCWOa	0.3	1.73	0.04	0	0	1.46
RDL2	0.29	1.91	0.31	0	0	0.92
CNCPos	0.28	0.22	0.87	0	0	0.38
PCVERBz	0.28	1.5	0.14	0	0	1.23
LSAGNd	0.28	1.86	0.19	0	0	1.07
CRFCWO1d	0.25	1.57	0.74	0	0	0
WRDFRQa	0.25	0.44	0.67	0	0	0.46
CNCCaus	0.25	0.36	0.29	0	0	1.15
CRFAOa	0.24	1.74	0.04	0	0	1.07
DESWLltd	0.23	0.2	0.49	0	0	0.69
WRDADJ	0.22	0.25	0.43	0	0	0.69
WRDPRP3p	0.21	1.25	0.51	0	0	0.23
LDMTLD	0.21	0.26	0.42	0	0	0.69
CRFCWOad	0.21	1.68	0.35	0	0	0.38
CNCTemp	0.2	0.55	0.06	0	1.45	0.15
WRDIMGc	0.19	0.4	0.22	0	0	0.84
DESWLsyd	0.19	0.69	0.26	0	0	0.69
WRDVERB	0.18	0.55	0.44	0	0	0.31
CRFNOa	0.15	1.06	0.07	0	0	0.61
CNCADC	0.14	1.46	0.12	0	0	0.31
WRDHYPv	0.14	0.35	0.32	0	0	0.31
WRDFRQc	0.14	0.58	0.38	0	0	0.15
WRDCNCc	0.14	0.37	0.51	0	0	0
LSASSp	0.14	1.54	0.08	0	0	0.38
LSASSpd	0.14	1.55	0.05	0	0	0.46
PCTEMPp	0.13	1.36	0.06	0	0	0.38
DRVP	0.12	0.92	0.15	0	0	0.31
WRDPRP2	0.12	1.54	0.13	0	0.23	0
DRGERUND	0.12	0.49	0.15	0	0	0.46
SMCAUSvp	0.11	0.54	0.28	0	0	0.15
PCSYNp	0.1	0.68	0.15	0	0	0.23
DRINF	0.1	0.68	0.11	0	0	0.31
DRAP	0.09	0.43	0.29	0	0	0
WRDADV	0.09	0.6	0.28	0	0	0
DESSLd	0.09	1	0.12	0	0.1	0.08
SYNSTRUTa	0.08	1.31	0.09	0	0	0
PCCNCp	0.07	1.21	0	0	0	0.15
WRDFAMc	0.07	0.63	0.2	0	0	0
SMCAUSv	0.06	0.53	0.14	0	0	0
CNCTempx	0.05	0	0.16	0	0	0.08
SMCAUSr	0.04	0.77	0.02	0	0	0
PCVERBp	0.04	0.9	0	0	0	0
WRDPRP1p	0.03	0.68	0.02	0	0	0
WRDPRP3s	0.02	0.4	0	0	0	0
DRNEG	0.01	0.24	0.02	0	0	0