-
Notifications
You must be signed in to change notification settings - Fork 0
/
STMLogAnalysis.Rmd
904 lines (726 loc) · 28 KB
/
STMLogAnalysis.Rmd
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
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
---
title: "CrystalBall Interaction Logs: STM Analysis"
author: "Ryan Wesslen"
date: "`r format(Sys.time(), '%d %B, %Y')`"
output:
html_document:
toc: true
toc_float: true
---
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo=TRUE, warning=FALSE, message=FALSE)
```
## Load the data
Load the data and convert to a data frame.
```{r data}
library(rjson)
logs <- fromJSON(file="./data/crystalball_userlog_final.json")
logs <- lapply(logs, function(x) {
x[sapply(x, is.null)] <- NA
unlist(x)
})
df <- do.call("rbind", logs)
df <- data.frame(user = row.names(df), log = df, stringsAsFactors = F)
```
How many unique users do we have?
```{r}
length(unique(df$user))
```
The data excludes test accounts and four users for incomplete data.
First, reshape the data from "wide" to "long" to create the time variables.
```{r}
library(reshape)
df <- reshape(df, varying = c("log.1","log.2","log.3","log.4","log.5",
"log.6","log.7","log.8","log.9","log.10"), idvar = "user", direction = "long")
```
Next, create two custom variables for our groups.
```{r}
df$frame <- ifelse(substr(df$user,0,1)=="h","High","Low")
df$train <- ifelse(substr(df$user,2,2)=="t","Time","Spatial")
table(df$frame, df$train)/10
df$time <- df$time / 10
```
Notice that the groups are spread fairly evenly.
### Appending User Attributes
First, upload user-level attributes.
```{r}
atts <- readr::read_csv("./data/demographics-pretest.csv")
```
Next, use the `pairsD3` function to scatter the personality scores.
```{r}
library(pairsD3)
pairsD3(atts[,9:13], group = as.factor(atts$GENDER))
```
### Personality K-Means
Run a quick k=5 k-means. This was explored as a feature for personality clusters.
```{r}
kmeans <- kmeans(atts[,9:13], centers = 5, iter.max = 10, nstart =3,
algorithm = c("Hartigan-Wong"), trace=FALSE)
pairsD3(atts[,9:13], group = as.factor(kmeans$cluster))
atts$flag2 <- ifelse(kmeans$cluster=="2",1,0)
atts$flag3 <- ifelse(kmeans$cluster=="3",1,0)
```
Alternatively, create binary flags for each personality measure.
Values above the mean are given a 1, values below are 0.
```{r}
summary(atts[,9:13])
atts$ExtravertFlag <- ifelse(atts$Extraversion>mean(atts$Extraversion),1,0)
atts$AgreeableFlag <- ifelse(atts$Agreeableness>mean(atts$Agreeableness),1,0)
atts$ConscientiousFlag <- ifelse(atts$Conscientiousness>mean(atts$Conscientiousness),1,0)
atts$NeuroticismFlag <- ifelse(atts$Neuroticism>mean(atts$Neuroticism),1,0)
atts$OpennessFlag <- ifelse(atts$Openness>mean(atts$Openness),1,0)
library(dplyr)
atts %>%
group_by(substr(atts$ID,1,2)) %>%
summarise(extra = mean(Extraversion),
agree = mean(Agreeableness),
consc = mean(Conscientiousness),
neuro = mean(Neuroticism),
open = mean(Openness))
```
We can test significant levels with a t-test.
```{r}
t.test(atts$Agreeableness~atts$GENDER)
```
### Major & Graduate Level
Last, let's create binary flags for major graduate level (Undegrad = 1, Grad/Prof = 0)
```{r}
table(atts$Occupation)
atts$Undergrad <- ifelse(atts$Occupation=="Undergraduate",1,0)
t.test(atts$Openness~atts$Undergrad)
```
Also, let's explore the major. I have manually created a flag (MajorS) for Computing and Non-Computing.
```{r}
table(atts$Major)
table(atts$MajorS)
df2 <- merge(df, atts, by.x = "user", by.y = "ID", all.x = TRUE)
```
## Log Preprocessing
Let's explore an example of the log sequence.
```{r}
df$log[1]
```
Note that most of the actions are divided into two words, with a "_". The one exception is "menu date". Let's correct them and then use a string split on "-".
```{r}
alt <- c("menu date picker_click","menu_date_picker_click",
"social network_hover","social_network_hover",
"social network_navigate","social_network_navigate",
"social network_click","social_network_click",
"word cloud_click","word_cloud_click",
"word cloud_hover","word_cloud_hover",
"word cloud_navigate","word_cloud_navigate",
"menu_click_menu icon","menu_click_menu_icon",
"menu_click_button followerfriend ratio","menu_click_button_followerfriend_ratio",
"map_click_button date distribution","map_click_button_date_distribution",
"menu_click_tweet icon","menu_click_tweet_icon",
"map_click_button find events","map_click_button_find_events",
"menu_click_favorite icon","menu_click_favorite_icon",
"followerfriend ratio","followerfriend_ratio",
"tweet_click_button more","tweet_click_button_more")
# replace all actions with _
for (i in c(1,3,5,7,9,11,13,15,17,19,21,23,25,27,29)){
df$log <- gsub(alt[1*i],alt[1*i+1],df$log)
}
# replace " - " with " "
df$log <- gsub(" - "," ",df$log)
df$log <- gsub("_",".",df$log)
df$log <- gsub(" "," ",df$log)
raw <- strsplit(df$log, " ")
df$actions <- as.integer(lapply(raw, length))
df$unique.actions <- as.integer(lapply(lapply(raw, unique), length))
write.csv(df, "./data/clean.csv", row.names = F)
```
Explore sample action statistics by group.
```{r}
library(tidyverse)
df %>% group_by(frame, train, time) %>% summarise(AvgActions=mean(actions),
StdActions=sd(actions),
AvgUnqActions=mean(unique.actions),
StdUnqActions=sd(unique.actions))
```
While at first glance it appears that High-Time and Low-Time have very high number of actions, notice that the standard deviations are very large -- so much so that, for example, that all the other values are within 1 std dev of the largest (High-Time). This means that there is not a large enough sample to derive much statistical significance.
We can plot the histograms.
```{r}
hist(df$actions,
breaks = 20,
main = "Histogram of Number of User Actions",
xlab = "User Actions",
ylab = "Number of Participant by Time Decile")
hist(df$unique.actions,
breaks = 20,
main = "Histogram of Number of Unique Actions",
xlab = "Unique Actions",
ylab = "Number of Participant by Time Decile")
```
Note that we are getting more distinct distributions by creating more documents (e.g. by breaking down the logs by time) and adding more distinct actions. Note that there are nearly 30 unique actions now.
## Bag-of-Actions Counts
Let's run a bag-of-words (DFM) with a 1-3 n-gram.
```{r}
library(quanteda)
myCorpus <- corpus(df$log)
docvars(myCorpus, "frame") <- df$frame
docvars(myCorpus, "train") <- df$train
docvars(myCorpus, "time") <- df$time
docvars(myCorpus, "extra") <- df2$ExtravertFlag
docvars(myCorpus, "agree") <- df2$AgreeableFlag
docvars(myCorpus, "conscient") <- df2$ConscientiousFlag
docvars(myCorpus, "neurotic") <- df2$NeuroticismFlag
docvars(myCorpus, "open") <- df2$OpennessFlag
docvars(myCorpus, "gender") <- df2$GENDER
docvars(myCorpus, "major") <- df2$MajorS
docvars(myCorpus, "undergrad") <- df2$Undergrad
docvars(myCorpus, "age") <- ifelse(df2$AGE>22,1,0)
# run DTM
dfm <- dfm(myCorpus, ngrams = 1:3)
# remove features not found in at least 50 "docs" -- modified from 25 when user-level
dfm <- dfm_trim(dfm, min_docfreq = 50)
```
Let's examine the top actions...
```{r}
x <- topfeatures(dfm)
```
```{r}
plot(tfidf(dfm))
```
We can also run a dendrogram (clustering). I left the labels off because their length (e.g. tri-actions) is too long.
```{r}
wordDfm <- sort(weight(dfm, "tf"))
wordDfm <- t(wordDfm)[1:50,] # keep the top 50 words
wordDistMat <- dist(wordDfm)
wordCluster <- hclust(wordDistMat)
# Define nodePar
plot(wordCluster, main="Frequency weighting (Labels Removed)", xlab=NA, sub=NA, labels = F)
```
## Structural Topic Modeling
Let's convert our data to the `stmdfm`.
```{r results="hide"}
library(stm)
# use quanteda converter to convert our Dfm
stmdfm <- convert(dfm, to = "stm", docvars = docvars(myCorpus))
out <- prepDocuments(stmdfm$documents, stmdfm$vocab, stmdfm$meta, lower.thresh = 50)
```
### Search for K
Let's first run a model search to obtain the number of topics.
```{r results="hide"}
c <- c(5,8,10,15,20,30,40)
kresult <- searchK(out$documents,
out$vocab,
K = c,
prevalence=~ s(time,5) + train + frame,
data=out$meta,
max.em.its = 100,
seed = 300
)
plot(kresult)
```
From this analysis, we decided on 8 topics given the high semantic coherence, relatively high held-out likelihood and parsimony. One key reason was we found 8 topics is where we found the "elbow" of the held-out likelihood. Note: we decided on 8 topics rather than 10 topics (e.g. since lower average Semantic Coherence) because the two additional topics simply broke out the Map View and Calendar View into to separate topics respectively (Map View Pan and Click, and Calendar View Dotted vs Solid Line). Collectively, we decided for the sake of parsimony, it would make more sense to keep these two views together rather than seperated.
We tested our model results (estimated anchor bias effects) with both (8 or 10 topics) and found both results were consistent.
### Run Baseline Model
Let's run the model with three covariates: time (b-spline), training and framing.
```{r results="hide"}
k <- 8
stmFit <- stm(out$documents, out$vocab, K = k, prevalence =~ s(time,5) + train + frame,
max.em.its = 150, data = out$meta, init.type = "Spectral", seed = 300)
```
```{r}
topicNames <- labelTopics(stmFit, n = 5)
t(topicNames$prob)
```
Using the word probabilities (top 5), let's give each topic the name of the first action (for simplicity).
```{r}
topic <- data.frame(
topicnames = c('Word Cloud: Navigation',
'Social Network: Navigation',
'Event List: All Tools',
'Social Network & Word Cloud',
'Word Cloud: Click',
'Calender View',
'Event Location & Flower Glyph',
'Map View'
),
TopicNumber = 1:k,
TopicProportions = colMeans(stmFit$theta))
```
Let's plot the topic sizes.
```{r}
plot(stmFit,
type = "summary",
xlim = c(0,.45),
ylim = c(0.2,8.4),
custom.labels = topic$topicnames,
main = "Topic Proportions by Interaction Clusters",
text.cex = 1)
```
Next, let's calculate the effect of the two factors (frame = High or Low, and train = Geo or Time).
```{r}
prep <- estimateEffect(1:k ~ s(time,5) + frame + train, stmFit, meta = out$meta, uncertainty = "Global")
```
### Effect of Visual Anchor
We can then use the `plot.estimateEffect` function to compare the effect of the "train" field (Geo or Time) on topic proportions (likelihood of the topic).
```{r train-plot, fig.height = 8, fig.width = 8, include=FALSE}
Result <- plot(prep, "train", method = "difference",
cov.value1 = "Time", cov.value2 = "Spatial",
verbose.labels = F,
custom.labels = topic$topicnames,
labeltype = "custom",
ylab = "Expected Difference in Topic Probability by Training (with 95% CI)",
xlab = "More Likely Geo Not Significant More Likely Time",
main = "Effect of Training on Topic Proportions",
xlim = c(-0.2,0.2))
```
We can then rank the topics by their point estimates.
```{r}
# order based on Expected Topic Proportion
trank = order(topic$TopicProportions, decreasing = T)
temp.topic <- topic[trank,]
x <- plot(prep, "train", method = "difference", model = stmFit,
cov.value1 = "Time", cov.value2 = "Spatial",
verbose.labels = F,
topics = temp.topic$TopicNumber,
custom.labels = temp.topic$topicnames,
#custom.labels = " ",
labeltype = "custom",
ylab = "Expected Topic Difference",
xlab = "Geo Not Significant Time",
main = "Effect of Visual Anchor",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
width = 40,
ci.level = 0.95)
```
### Effect of Numeric Anchor
```{r frame-plot, fig.height = 8, fig.width=8, include=FALSE}
Result <- plot(prep, "frame", method = "difference",
cov.value1 = "High", cov.value2 = "Low",
verbose.labels = F,
custom.labels = topic$topicnames,
labeltype = "custom",
ylab = "Expected Difference in Topic Probability by Framing (with 95% CI)",
xlab = "More Likely Low Not Significant More Likely High",
main = "Effect of Framing on Topic Proportions",
xlim = c(-0.2,0.2),
width = 40)
```
```{r}
# optional: order based on Topic Prevalance
frank = order(unlist(Result$means))
#temp.topic <- topic[frank,]
plot(prep, "frame", method = "difference", model = stmFit,
cov.value1 = "High", cov.value2 = "Low",
verbose.labels = F,
topics = temp.topic$TopicNumber,
custom.labels = temp.topic$topicnames,
#custom.labels = " ",
labeltype = "custom",
ylab = "Expected Topic Difference (95% CI)",
xlab = "Low Not Significant High",
main = "Effect of Numerical Anchor",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
width = 40,
ci.level = 0.95)
```
### Effect of Time
95% confidence intervals have been added.
```{r time-plot}
par(mfrow = c(2,2),mar = c(4,4,2,2))
for (i in trank){
plot(prep, "time", method = "continuous", topics = i, model = z,
main = paste0(topic$topicnames[i]),
printlegend = FALSE, ylab = "Exp. Topic Prob",
xlab = "Time (Deciles of Action)", ylim = c(-0.01,0.5),
ci.level = 0.95
)
}
```
### Interaction: Time & Visual Anchor
```{r finteraction}
par(mfrow = c(2,2),mar = c(4,4,2,2))
for (i in trank){
plot(prep, "time", method = "continuous", topics = i, main = paste0(topic$topicnames[i]),
printlegend = FALSE, ylab = "Exp. Topic Prob", xlab = "Time (Deciles of Action)",
moderator = "train", moderator.value = "Time", linecol = "red", ylim = c(-0.01 ,0.5),
ci.level = 0)
plot(prep, "time", method = "continuous", topics = i,
printlegend = FALSE, ylab = "Exp. Topic Prob", xlab = "Time (Deciles of Action)",
moderator = "train", moderator.value = "Spatial", linecol = "blue", add = "T",
ylim = c(-0.01 ,0.5), ci.level = 0)
legend(0.65, 0.5, c("Time", "Geo"), lwd = 2, col = c("red", "blue"))
}
```
### Interaction: Time & Numeric Anchor
```{r interaction}
par(mfrow = c(2,2),mar = c(4,4,2,2))
for (i in frank){
plot(prep, "time", method = "continuous", topics = i, main = paste0(topic$topicnames[i]),
printlegend = FALSE, ylab = "Exp. Topic Prob", xlab = "Time (Deciles of Action)",
moderator = "frame", moderator.value = "High", linecol = "green", ylim = c(-0.01 ,0.5),
ci.level = 0)
plot(prep, "time", method = "continuous", topics = i,
printlegend = FALSE, ylab = "Exp. Topic Prob", xlab = "Time (Deciles of Action)",
moderator = "frame", moderator.value = "Low", linecol = "brown", add = "T",
ylim = c(-0.01 ,0.5), ci.level = 0)
legend(0.65, 0.5, c("Low", "High"), lwd = 2, col = c("brown", "green"))
}
```
## Personality Covariates
Let's keep time and training and used different binary flags.
```{r include=FALSE}
stmFit2 <- stm(out$documents, out$vocab, K = k,
prevalence =~ s(time,5) + train + extra + agree + conscient + neurotic + open,
max.em.its = 150, data = out$meta, init.type = "Spectral", seed = 300)
```
```{r}
topicNames2 <- labelTopics(stmFit2)
topic2 <- data.frame(
topicnames = c('Word Cloud: Navigation',
'Social Network: Navigation',
'Event List: All Tools',
'Social Network & Word Cloud',
'Word Cloud: Click',
'Calender View',
'Event Location & Flower Glyph',
'Map View'
),
TopicNumber = 1:k,
TopicProportions = colMeans(stmFit2$theta))
```
```{r, fig.height=6, fig.width=8, include=FALSE}
plot(stmFit2,
type = "summary",
xlim = c(0,.5),
n =1,
main = "Log (Action) Topics",
text.cex = 1)
```
```{r, include=FALSE}
prep2 <- estimateEffect(1:k ~ s(time,5) + train + extra + agree + conscient + neurotic + open,
stmFit2, meta = out$meta, uncertainty = "Global")
```
### Effect of Extravert
```{r extra}
extraResult <- plot(prep2, "extra", method = "difference",
cov.value1 = "1", cov.value2 = "0",
verbose.labels = F,
model = stmFit2,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Low Not Significant High",
main = "Effect of Extraversion",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
### Effect of Agreeable
```{r agree}
agreeResult <- plot(prep2, "agree", method = "difference",
cov.value1 = "1", cov.value2 = "0",
verbose.labels = F,
model = stmFit2,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Low Not Significant High",
main = "Effect of Agreeableness",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
### Effect of Conscientious
```{r consc}
consResult <- plot(prep2, "conscient", method = "difference",
cov.value1 = "1", cov.value2 = "0",
verbose.labels = F,
model = stmFit2,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Low Not Significant High",
main = "Effect of Conscientiousness",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
### Effect of Neurotic
```{r neurotic}
neuroResult <- plot(prep2, "neurotic", method = "difference",
cov.value1 = "1", cov.value2 = "0",
verbose.labels = F,
model = stmFit2,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Low Not Significant High",
main = "Effect of Neuroticism",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
### Effect of Openness
```{r open}
openResult <- plot(prep2, "open", method = "difference",
cov.value1 = "1", cov.value2 = "0",
verbose.labels = F,
model = stmFit2,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Low Not Significant High",
main = "Effect of Openness",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
### Combined Effect
This matrix scatter plot shows the values of the estimated effects for each topic by each of the five binary variables.
```{r}
df3 <- data.frame(label = unlist(agreeResult$labels),
Agreeable = unlist(agreeResult$means),
Conscientious = unlist(consResult$means),
Extravert = unlist(extraResult$means),
Neurotic = unlist(neuroResult$means),
Openness = unlist(openResult$means))
# another option
makePairs <- function(data)
{
grid <- expand.grid(x = 1:ncol(data), y = 1:ncol(data))
grid <- subset(grid, x != y)
all <- do.call("rbind", lapply(1:nrow(grid), function(i) {
xcol <- grid[i, "x"]
ycol <- grid[i, "y"]
data.frame(xvar = names(data)[ycol], yvar = names(data)[xcol],
x = data[, xcol], y = data[, ycol], data)
}))
all$xvar <- factor(all$xvar, levels = names(data))
all$yvar <- factor(all$yvar, levels = names(data))
densities <- do.call("rbind", lapply(1:ncol(data), function(i) {
data.frame(xvar = names(data)[i], yvar = names(data)[i], x = data[, i])
}))
list(all=all, densities=densities)
}
# expand iris data frame for pairs plot
gg1 = makePairs(df3[,2:6])
# pairs plot
# https://gastonsanchez.wordpress.com/2012/08/27/scatterplot-matrices-with-ggplot/
ggplot(gg1$all, aes_string(x = "x", y = "y")) +
facet_grid(xvar ~ yvar, scales = "free") +
xlim(-0.05,0.05) + ylim(-0.05,0.05) +
geom_point( na.rm = TRUE, alpha=0.8) +
#geom_point(aes(colour=Species), na.rm = TRUE, alpha=0.8) +
stat_density(aes(x = x, y = ..scaled.. * diff(range(x)) + min(x)),
data = gg1$densities, position = "identity",
colour = "grey20", geom = "line")
```
## Gender Covariate
Let's run Male/Female as a covariate.
```{r include = FALSE}
stmFit3 <- stm(out$documents, out$vocab, K = k, prevalence =~ s(time,5) + train + gender,
max.em.its = 150, data = out$meta, init.type = "Spectral", seed = 300)
```
```{r, include=FALSE}
topicNames <- labelTopics(stmFit3)
topic <- data.frame(
topicnames = c('Word Cloud: Navigation',
'Social Network: Navigation',
'Event List: All Tools',
'Social Network & Word Cloud',
'Word Cloud: Click',
'Calender View',
'Event Location & Flower Glyph',
'Map View'
),
TopicNumber = 1:k,
TopicProportions = colMeans(stmFit3$theta))
```
```{r, fig.height=6, fig.width=8, include=FALSE}
plot(stmFit3,
type = "summary",
xlim = c(0,.3),
n =1,
main = "Log (Action) Topics",
text.cex = 1)
```
```{r, include=FALSE}
prep3 <- estimateEffect(1:k ~ s(time,5) + train + gender, stmFit3, meta = out$meta, uncertainty = "Global")
```
### Effect of Gender
```{r gender}
Result <- plot(prep3, "gender", method = "difference",
cov.value1 = "F", cov.value2 = "M",
verbose.labels = F,
model = stmFit3,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Male Not Significant Female",
main = "Effect of Gender",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
Result: No Significant Actions (Topics)
## Education Level Covariate
We'll use a binary flag for undergraduates (1) and graduate students (0).
```{r include = FALSE}
stmFit4 <- stm(out$documents, out$vocab, K = k, prevalence =~ s(time,5) + train + undergrad,
max.em.its = 150, data = out$meta, init.type = "Spectral", seed = 300)
```
```{r include=FALSE}
topicNames <- labelTopics(stmFit4)
topic <- data.frame(
topicnames = c('Word Cloud: Navigation',
'Social Network: Navigation',
'Event List: All Tools',
'Social Network & Word Cloud',
'Word Cloud: Click',
'Calender View',
'Event Location & Flower Glyph',
'Map View'
),
TopicNumber = 1:k,
TopicProportions = colMeans(stmFit4$theta))
```
```{r, fig.height=6, fig.width=8, include=FALSE}
plot(stmFit4,
type = "summary",
xlim = c(0,.3),
n =1,
main = "Log (Action) Topics",
text.cex = 1)
```
```{r, include=FALSE}
prep4 <- estimateEffect(1:k ~ s(time,5) + train + undergrad, stmFit4, meta = out$meta, uncertainty = "Global")
```
### Effect of Education Level
```{r undergrad}
Result <- plot(prep4, "undergrad", method = "difference",
cov.value1 = "0", cov.value2 = "1",
verbose.labels = F,
model = stmFit4,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Undergrad Not Significant Graduate",
main = "Effect of Education Level",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
## Major Covariate
We'll use a binary flag for major: computing-related (1) and non-computing (social sciences and humanities) (0).
```{r include = FALSE}
stmFit5 <- stm(out$documents, out$vocab, K = k, prevalence =~ s(time,5) + train + major,
max.em.its = 150, data = out$meta, init.type = "Spectral", seed = 300)
```
```{r include=FALSE}
topicNames <- labelTopics(stmFit5)
topic <- data.frame(
topicnames = c('Word Cloud: Navigation',
'Social Network: Navigation',
'Event List: All Tools',
'Social Network & Word Cloud',
'Word Cloud: Click',
'Calender View',
'Event Location & Flower Glyph',
'Map View'
),
TopicNumber = 1:k,
TopicProportions = colMeans(stmFit5$theta))
```
```{r include=FALSE}
plot(stmFit5,
type = "summary",
xlim = c(0,.3),
n =1,
main = "Log (Action) Topics",
text.cex = 1)
```
```{r include=FALSE}
prep5 <- estimateEffect(1:k ~ s(time,5) + train + major, stmFit5, meta = out$meta, uncertainty = "Global")
```
### Effect of Major
```{r major}
Result <- plot(prep5, "major", method = "difference",
cov.value1 = "0", cov.value2 = "1",
verbose.labels = F,
model = stmFit5,
labeltype = "custom",
custom.labels = topic2$topicnames,
#custom.labels = " ",
ylab = "Exp Topic Difference",
xlab = "Computing Not Significant Non-Computing",
main = "Effect of Major",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
## Age Covariate
We'll use a binary flag for age: over 23 or older (1) or under 23 (0).
```{r include = FALSE}
stmFit5 <- stm(out$documents, out$vocab, K = k, prevalence =~ s(time,5) + train + age,
max.em.its = 150, data = out$meta, init.type = "Spectral", seed = 300)
```
```{r include=FALSE}
topicNames <- labelTopics(stmFit5)
topic <- data.frame(
topicnames = c('Word Cloud: Navigation',
'Social Network: Navigation',
'Event List: All Tools',
'Social Network & Word Cloud',
'Word Cloud: Click',
'Calender View',
'Event Location & Flower Glyph',
'Map View'
),
TopicNumber = 1:k,
TopicProportions = colMeans(stmFit5$theta))
```
```{r include=FALSE}
plot(stmFit5,
type = "summary",
xlim = c(0,.3),
n =1,
main = "Log (Action) Topics",
text.cex = 1)
```
```{r include=FALSE}
prep5 <- estimateEffect(1:k ~ s(time,5) + train + age, stmFit5, meta = out$meta, uncertainty = "Global")
```
### Effect of Age
```{r age}
Result <- plot(prep5, "age", method = "difference",
cov.value1 = "1", cov.value2 = "0",
verbose.labels = F,
model = stmFit5,
custom.labels = topic$topicnames,
#custom.labels = " ",
labeltype = "custom",
ylab = "Exp Topic Difference",
xlab = "18-22 Not Significant 23+ Older",
main = "Effect of Age",
xlim = c(-0.2,0.2),
ylim = c(0.4,8.4),
ci.level = 0.95,
width = 40)
```
## Save Image and Packages Used
```{r}
save.image(file = "./data/stmimage.Rdata")
sessionInfo()
```