-
Notifications
You must be signed in to change notification settings - Fork 0
/
git_script_SC_V1.0.R
275 lines (213 loc) · 15.8 KB
/
git_script_SC_V1.0.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
# R-ANALYSIS SCRIPT FOR
#
# If you use this code for analysis that is published in an indexed journal or repository, please cite the following article:
#
# Janssen, N. & Lopéz-Peréz, P. J., Determinants of response latencies in the sentence completion task.
#
# Niels Janssen, Sept. 2017.
rm(list=ls(all=TRUE))
setwd('set/to/path') # put correct path here!
dt = read.table("data_matrix_SC_v1.0.txt", header = TRUE)
nrow(dt) # = 5703 total cleaned trials
####
# STEP 1 - new variables
####
# 1. Frequency of the response (LexFreq, here: logLexespC)
# 2. Length of the response (RLen, here: num_phonC)
# 3. Length of the sentence (SLen, here: slenC)
# 4. Phonological neighborhood of the response (PNS, here: logPNC)
# 5. Lexical Diversity of the participant (LexDiv, here: scohortC)
# 6. Response counts (RespCounts, here: logcountsC)
# 7. Sentence Cohort size (CohortSize, here: cohortsizeC)
# 8. Articulatory 1 (Plosive, here: init_phonC)
# 9. Articulatory 2 (Voiced, here: voicedC)
#
# Strategy here is to for try to determine the by subject random slopes, and leave the by-item slopes empty
# then after getting the minimal model for by-subjects, start with a full mode by-items and trim that down.
# This avoids starting with a full model that does not converge-
require(lme4)
# Do Bates et al. (2015) first by-subjects
model1s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + voicedC + init_phonC||subject) + (1|item), data = dt, REML = FALSE)
summary(model1s)
# remove by-subject num_phonC, loglexespC
model2s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+cohortsizeC + logcountsC + logPNC + slenC + voicedC + init_phonC||subject) + (1|item), data = dt, REML = FALSE)
anova(model1s,model2s) # not-significant, so simpler model2s justified.
summary(model2s)
# remove by-subject cohortsize
model3s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+logcountsC + logPNC + slenC + voicedC + init_phonC||subject) + (1|item), data = dt, REML = FALSE)
anova(model2s,model3s) # not-significant, so simpler model3s justified.
summary(model3s)
# remove by-subject slenC
model4s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+logcountsC + logPNC + voicedC + init_phonC||subject) + (1|item), data = dt, REML = FALSE)
anova(model3s,model4s) # not-significant, so simpler model3s justified.
summary(model4s)
# remove by-subject voicedC
model5s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+logcountsC + logPNC + init_phonC||subject) + (1|item), data = dt, REML = FALSE)
anova(model4s,model5s) # not-significant, so simpler model3s justified.
summary(model5s)
# remove by-subject logPNC
model6s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+logcountsC + init_phonC||subject) + (1|item), data = dt, REML = FALSE)
anova(model5s,model6s) # not-significant, so simpler model3s justified.
summary(model6s)
# remove by-subject init_phonC
model7s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+ logcountsC||subject) + (1|item), data = dt, REML = FALSE)
anova(model6s,model7s) # not-significant, so simpler model3s justified.
summary(model7s)
# remove by-subject logcountsC
model8s = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1|item), data = dt, REML = FALSE)
anova(model7s,model8s) # significant! model7s is final model
summary(model8s)
####
# same thing with by items
model1i = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1+ logcountsC + loglexespC + logPNC + num_phonC + scohortC + voicedC + init_phonC||item), data = dt, REML = FALSE)
summary(model1i)
# remove by-item scohort
model2i = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1+ logcountsC + loglexespC + logPNC + num_phonC + voicedC + init_phonC||item), data = dt, REML = FALSE)
anova(model1i,model2i) # not significant, removal justified.
summary(model2i)
# remove by-item loglexespC
model3i = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1+ logcountsC + logPNC + num_phonC + voicedC + init_phonC||item), data = dt, REML = FALSE)
anova(model2i,model3i) # not significant, removal justified.
summary(model3i)
# remove by-item logPNC
model4i = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1+ logcountsC + num_phonC + voicedC + init_phonC||item), data = dt, REML = FALSE)
anova(model3i,model4i) # not significant, removal justified.
summary(model4i)
# remove by-item init_phonC
model5i = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1+ logcountsC + num_phonC + voicedC ||item), data = dt, REML = FALSE)
anova(model4i,model5i) # not significant, removal justified.
summary(model5i)
# remove by-item voicedC
model5i = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE)
anova(model4i,model5i) # not significant, removal justified.
summary(model5i)
# remove by-item num_phonC or logcountsC
model6i = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1|subject) + (1+ logcountsC + num_phonC||item), data = dt, REML = FALSE)
anova(model5i,model6i) # significant! removal NOT justified. Final model is 5i.
summary(model5i)
###
# combine final models from subjects and items
# repeat procedure for fixed effects
model1 = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + logPNC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+ logcountsC||subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE)
summary(model1)
# remove fixed effect logPNC
model2 = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + num_phonC + slenC + scohortC + voicedC + init_phonC + (1+ logcountsC||subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE)
anova(model1, model2) # not signiticant, removal justified.
summary(model2)
# remove fixed effect voicedC
model3 = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + num_phonC + slenC + scohortC + init_phonC + (1+ logcountsC||subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE)
anova(model2, model3) # not signiticant, removal justified.
summary(model3)
# remove fixed effect num_phonC
model4 = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logcountsC||subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE)
anova(model3, model4) # not signiticant, removal justified.
summary(model4)
# remove fixed effect slenC
model5 = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + scohortC + init_phonC + (1+ logcountsC||subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE)
anova(model4, model5) # signiticant! removal NOT justified. model4 is final model.
summary(model5)
# testing for removal of significant effects in final model
model4a = lmer(logRT ~ cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logcountsC||subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE)
anova(model4, model4a)
# test for accumulation
anova(model1, model4)
anova(model1s, model7s)
anova(model1i, model5i)
# check condition index
source("mer-utils.R") # from Florian Jaeger website
kappa.mer(model4) # condition index is 2.00
# get p-values
require(afex)
mixed(logRT ~ cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logcountsC||subject) + (1+ logcountsC + num_phonC ||item), data = dt, REML = FALSE, method = "S")
####
# STEP 2 - existing variables
####
# 1. Cloze Probability (here: logagreeC)
# 2. Response Ratio (here: logstsratioC)
# 3. LSA Strength (here: logLSAC)
# 4. LSA Entropy (here: loglsaentC)
# effect of cloze-prob by itself
model4a1 = lmer(logRT ~ logagreeC + (1+ logagreeC|subject) + (1+ logagreeC |item), data = dt, REML = FALSE)
model4a2 = lmer(logRT ~ logagreeC + (1+ logagreeC|subject) + (1|item), data = dt, REML = FALSE)
mixed(logRT ~ logagreeC + (1+ logagreeC|subject) + (1|item), data = dt, REML = FALSE, method="S")
anova(model4a1,model4a2)
summary(model4a1)
# effect of cloze-prob with new vars
model4a = lmer(logRT ~ logagreeC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logagreeC+ logcountsC||subject) + (1+ logagreeC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logagreeC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logagreeC+ logcountsC||subject) + (1+ logagreeC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4a)
# effect of response ratio by itself
model4b1 = lmer(logRT ~ logstsratioC+ (1+ logstsratioC|subject) + (1+ logstsratioC |item), data = dt, REML = FALSE)
model4b2 = lmer(logRT ~ logstsratioC+ (1|subject) + (1|item), data = dt, REML = FALSE)
mixed(logRT ~ logstsratioC+ (1|subject) + (1|item), data = dt, REML = FALSE, method="S")
anova(model4b1,model4b2)
summary(model4b1)
# effect of response ratio
model4b = lmer(logRT ~ logstsratioC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logstsratioC+ logcountsC||subject) + (1+ logstsratioC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logstsratioC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logstsratioC+ logcountsC||subject) + (1+ logstsratioC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4b)
# effect of lsa strength by itself
model4c1 = lmer(logRT ~ logLSAC + (1+ logLSAC|subject) + (1+ logLSAC|item), data = dt, REML = FALSE)
model4c2 = lmer(logRT ~ logLSAC + (1+ logLSAC|subject) + (1|item), data = dt, REML = FALSE)
mixed(logRT ~ logLSAC + (1+ logLSAC|subject) + (1|item), data = dt, REML = FALSE, method="S")
anova(model4c1,model4c2)
summary(model4c1)
# effect of lsa strength
model4c = lmer(logRT ~ logLSAC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logLSAC+ logcountsC||subject) + (1+ logLSAC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logLSAC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logLSAC+ logcountsC||subject) + (1+ logLSAC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4c)
# effect of lsa ent by itself
model4d1 = lmer(logRT ~ loglsaentC + (1+ loglsaentC|subject) + (1+ loglsaentC|item), data = dt, REML = FALSE)
model4d2 = lmer(logRT ~ loglsaentC + (1+ loglsaentC|subject) + (1|item), data = dt, REML = FALSE)
mixed(logRT ~ loglsaentC + (1+ loglsaentC|subject) + (1+ loglsaentC|item), data = dt, REML = FALSE, method="S")
anova(model4d1, model4d2)
summary(model4d1)
# effect of lsa ent
model4d = lmer(logRT ~ loglsaentC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ loglsaentC+ logcountsC||subject) + (1+ loglsaentC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ loglsaentC + cohortsizeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ loglsaentC+ logcountsC||subject) + (1+ loglsaentC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4d)
####
# Examining why no effects of the four vars when combined with new variables
# Look at correlation matrix
####
corvars = c("logagreeC","logstsratioC","logLSAC","loglsaentC","cohortsizeC", "logcountsC", "loglexespC", "slenC", "scohortC")
cormat = round(cor(dt[,corvars]),2)
cormat # note high correlations between existing vars and response counts and cohort size
####
# Leaving out Cohort Size
####
# effect of cloze-prob, leaving out cohort size
model4a = lmer(logRT ~ logagreeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logagreeC+ logcountsC||subject) + (1+ logagreeC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logagreeC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logagreeC+ logcountsC||subject) + (1+ logagreeC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4a)
# effect of response ratio, leaving out cohort size
model4b = lmer(logRT ~ logstsratioC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logstsratioC+ logcountsC||subject) + (1+ logstsratioC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logstsratioC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logstsratioC+ logcountsC||subject) + (1+ logstsratioC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4b)
# effect of lsa strength, leaving out cohort size
model4c = lmer(logRT ~ logLSAC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logLSAC+ logcountsC||subject) + (1+ logLSAC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logLSAC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ logLSAC+ logcountsC||subject) + (1+ logLSAC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4c)
# effect of lsa ent, leaving out cohort size
model4d = lmer(logRT ~ loglsaentC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ loglsaentC+ logcountsC||subject) + (1+ loglsaentC + logcountsC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ loglsaentC + logcountsC + loglexespC + slenC + scohortC + init_phonC + (1+ loglsaentC+ logcountsC||subject) + (1+ loglsaentC + logcountsC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4d)
####
# Leaving out response counts
####
# effect of cloze-prob, leaving out response counts
model4a = lmer(logRT ~ logagreeC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ logagreeC||subject) + (1+ logagreeC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logagreeC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ logagreeC|subject) + (1+ logagreeC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4a)
# effect of response ratio, leaving out response counts
model4b = lmer(logRT ~ logstsratioC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ logstsratioC||subject) + (1+ logstsratioC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logstsratioC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ logstsratioC||subject) + (1+ logstsratioC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4b)
# effect of lsa strength, leaving out response counts
model4c = lmer(logRT ~ logLSAC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ logLSAC||subject) + (1+ logLSAC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ logLSAC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ logLSAC||subject) + (1+ logLSAC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4c)
# effect of lsa ent, leaving out response counts
model4d = lmer(logRT ~ loglsaentC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ loglsaentC||subject) + (1+ loglsaentC + num_phonC ||item), data = dt, REML = FALSE)
mixed(logRT ~ loglsaentC + cohortsizeC + loglexespC + slenC + scohortC + init_phonC + (1+ loglsaentC||subject) + (1+ loglsaentC + num_phonC ||item), data = dt, REML = FALSE, method="S")
summary(model4d)