Miroo Lee 12/14/2021
- Part 1. Cleaning the praat
output
- 1. Determine the encoding
- 2. Add proficiency level
- 3. Add list of words and their lexical stress info
- 4. Specify syllables’ positions in words and mark if stressed
- 5. Normalize segment durations
- 6. Clean the phone labels
- 7. Get word-initial syllables for DIS
- 8. Get word-final syllables for DFL
- 9. Get CVC+CV syllable data
- Part 2. Statistical Analysis on DIS on CV
After running a praat script that exports annotated information from Praat text grids, we get a txt file that needs cleaning. This document illustrates how I cleaned and prepared the data for the analysis.
I used a very handy guess_encoding function form readr package.
wav_folder <- "/Users/miroolee/Documents/DataScience/L2-Prosody-Analysis/wav_SAMPLES"
dat_path <- paste(wav_folder, "/duration_results.txt", sep="")
guess_encoding(dat_path)## # A tibble: 3 × 2
## encoding confidence
## <chr> <dbl>
## 1 UTF-16BE 1
## 2 windows-1250 0.39
## 3 windows-1252 0.35
dat <- read_tsv(dat_path, locale=locale(encoding="UTF-16"))## New names:
## * `` -> ...13
## Rows: 9333 Columns: 13
## ── Column specification ────────────────────────────────────────────────────────
## Delimiter: "\t"
## chr (8): Filename, PrecedingPhone, PhoneLabel, FollowingPhone, PitchMaxInPho...
## dbl (4): StartingPoint, PhoneDuration, SyllDuration, WordDuration
## lgl (1): ...13
##
## ℹ Use `spec()` to retrieve the full column specification for this data.
## ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
I united some parts from the previous PELIC output (1.3) to praat output
folder <- "/Users/miroolee/Documents/DataScience/L2-Prosody-Analysis"
lev_path <- paste(folder, "/KOR_mono.csv", sep="")
lev <- read.csv(lev_path)
lev1<-lev %>%
unite(Filename, c("row_id", "anon_id")) %>%
select (c(Filename, gender, level_id))
dat1 <- dat %>%
left_join(lev1, dat, by="Filename")For this part, I made a separate document containing syllable structure
and lexical stress information of all the words in the wav files
wordList.csv which can be found in my github
repository. This information was handcoded by me, and therefore needs to
be updated continuously as you add more wav files with new words.
Obviously this is not optimal, and I hope to replace this document with
some sort of dictionary later that contains syllable structure and
stress information.
Here I have three newly annotated wav files. I am combining the new words to my previous ‘wordList.csv’ file.
wdDat_path <- paste(folder, "/wordList.csv", sep="")
wdDat <- read.csv(wdDat_path)
dat2 <- dat1 %>%
left_join(wdDat, by =c("SyllLabel" = "SyllLabel", "WordLabel" = "WordLabel"))
new_wdDat <- dat2 %>%
filter(Filename=="21819_ea4"|Filename=="23027_ea4"|Filename=="24473_ea4") %>%
distinct(WordLabel,SyllLabel, .keep_all=TRUE) #I think I can also use anti_join() but oh well.
write.csv(new_wdDat, 'new_wordList.csv')Next step is to work on new_wordList.csv that does not have values in SyllCV, primary and secondary stress. This is the part that needs handcoding (until I figure out how to replace it with dictionary). For now, I will work with dat2.
I added two new variables:
[SyllOrder] has with 3 levels: (wd) initial, medial and final.
[stress] has 2 levels: stressed, unstressed.
There’s an issue in the PrimaryStress coding. Now if syllable 1,2,3 are stressed (ex:USA), PrimaryStress value shows as 123. This would be shown as ‘unstressed’ in stress column. The number of this exception is not great, and we will move forward with the data as it is for now.
dat3<-dat2 %>%
select(-c(PitchMaxInPhone,...13)) %>%
separate(SyllLabel, into = c("currentSyll", "entireSyll", sep ="_" )) %>%
add_column(SyllOrder = "medial") %>%
mutate (SyllOrder = case_when(
currentSyll == "1" ~ "initial",
currentSyll == entireSyll ~ "final",
currentSyll < entireSyll ~ "medial")) %>%
mutate (stress = case_when(
currentSyll == PrimaryStress ~ "stressed",
currentSyll != PrimaryStress ~ "unstressed"))## Warning: Expected 3 pieces. Additional pieces discarded in 1 rows [1].
## Warning: Expected 3 pieces. Missing pieces filled with `NA` in 9332 rows [2, 3,
## 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, ...].
I added one new variable: [normedPhoneDur] which is calculated by dividing raw duration by the syllable duration. This was used specifically for voiceless stops VOT and voiceless fricatives to account for speech rate.
dat4 <- dat3 %>%
mutate(normedPhoneDur = PhoneDuration/SyllDuration )Removed numbers from [PrecedingPhone].[PhoneLabel].[FollowingPhone].
dat5 <- dat4 %>%
mutate_at("PrecedingPhone", str_replace, "[12345]_", "") %>%
mutate_at("PhoneLabel", str_replace, "[12345]_", "") %>%
mutate_at("FollowingPhone", str_replace, "[12345]_", "")In this process, I separated word-initial syllables for the analysis of DIS (Domain Initial Strengthening) effect. Similarly, I separated word-final syllables for the analysis of DFL (Domain Fianl Lengthening) effect.
Note that DIS is operationalized as preceding pause duration as a proxy measure for the strength of prosodic domain junctures. I assumes longer the pause is, higher the prosodic domain is. I expect stronger DIS for longer pause.
I selected word-initial syllables, and then assigned preceding pause level for each word-initial syllable using a new variable [DIS]. All the word-initial syllables were saved as ‘firstSyll_dat’.
firstSyll_dat <- dat5 %>%
group_by(WordLabel,SyllDuration) %>%
filter(currentSyll==1)%>%
mutate(DIS = case_when(
PrecedingPhone == "L" ~ "3",
PrecedingPhone == "M" ~ "2",
PrecedingPhone == "S" ~ "1")) %>%
fill(DIS, .direction="down") %>%
replace_na(list(DIS = 0)) %>%
mutate(SyllCV=factor(SyllCV))I selected word-final syllables and assigned following pause level for each word-final syllable using a new variable [DFL], short for domain_final_lengthening. All the word-beignning syllables were saved as a separate dataframe named ‘finSyll_dat’.
finSyll_dat <- dat5 %>%
group_by(WordLabel,SyllDuration) %>%
filter(currentSyll == entireSyll) %>%
mutate(DFL = case_when(
FollowingPhone == "L" ~ "3",
FollowingPhone == "M" ~ "2",
FollowingPhone == "S" ~ "1")) %>%
fill(DFL, .direction="up") %>%
replace_na(list(DFL = 0)) %>%
mutate(SyllCV=factor(SyllCV))I made a subset of dataset of with the two most frequent syllable structures, CV and CVC. Note that the main interests in this analysis is any durational changes in vowels and voiceless consonants as a function of stress, preceding/following pause, and proficiency level.
Because VOT durations are differentially enhanced depending on a voicing feature(shorter for voiced, longer for voiceless), I only looked at voiceless consonants. For the similar reason, I only looked at single consonant onset, rather than looking at syllable onset with double or more consonants (ex: steam).
summary(firstSyll_dat$SyllCV)## V VC CV VCC CVV CVCC CCV CVC CVVC CVVCC CVVCCC
## 87 598 1142 255 267 571 66 1523 225 41 12
## CCVC CCVCC CVCCC CCVV VV VVC CCVVC CCCVC CVCCV NA's
## 107 78 22 2 144 23 45 6 3 1168
summary(finSyll_dat$SyllCV)## V CVC CVVC CCVC CV VC CVCC VCC CVV CCVCC CVVCC
## 41 1954 333 124 941 524 909 255 189 73 60
## CVVCCC CCV CVCCC VCV CCVV VV VVC CCVVC CCCVC CVCCV NA's
## 12 41 34 9 2 148 25 45 6 3 1759
Word_initial and word_final datasets were separately analyzed. Four
output datasets are made:
both_ini_V : vowels in word_initial syllable CVC+CV
both_ini_C: voiceless stops and fricatives in word_initial syllable
CVC+CV
both_fin_V: vowels in word_final syllable CVC+CV
both_fin_C: voiceless stops and fricatives in word_final syllable
CVC+CV
both_dat_ini <- firstSyll_dat %>%
filter(SyllCV %in% c("CVC","CV")) %>%
filter(PhoneLabel %in% c("t_vot","k_vot","p_vot","f","s","ð","ʃ","V")) %>%
mutate(DIS=factor(DIS)) %>%
mutate(DIS=fct_relevel(DIS,c("0","1","2","3"))) %>%
mutate(stress=factor(stress))%>%
mutate(stress=fct_relevel(stress,c("unstressed","stressed"))) %>%
mutate(level_id=factor(level_id))%>%
drop_na(level_id)
both_ini_V <- both_dat_ini %>%
filter(PhoneLabel %in% c("V"))
both_ini_C <- both_dat_ini %>%
filter(PhoneLabel %in% c("t_vot","k_vot","p_vot","f","s","ð","ʃ"))
both_dat_fin <- finSyll_dat %>%
filter(SyllCV %in% c("CVC","CV")) %>%
filter(PhoneLabel %in% c("t_vot","k_vot","p_vot","f","s","ð","ʃ","V")) %>%
mutate(DFL=factor(DFL)) %>%
mutate(DFL=fct_relevel(DFL,c("0","1","2","3"))) %>%
mutate(stress=factor(stress))%>%
mutate(stress=fct_relevel(stress,c("unstressed","stressed"))) %>%
mutate(level_id=factor(level_id))%>%
drop_na(level_id)
both_fin_V <- both_dat_fin %>%
filter(PhoneLabel %in% c("V"))
both_fin_C <- both_dat_fin %>%
filter(PhoneLabel %in% c("t_vot","k_vot","p_vot","f","s","ð","ʃ"))I tested how the word-initial syllable’s consonants(C)’ and vowels(V)’ durations are affected by the preceding 1.pause duration(0-3), the presence of 2.lexical stress(str vs unstr), and the learners’ 3.proficiency level(3 vs.5). As for the dependent variable, log([PhoneDuration]) was used for vowels and [normedPhonDur] for consonants.
lm1 = lmer(log(PhoneDuration)~DIS+stress+level_id+(1|WordLabel)+(1|Filename), data=both_ini_V)
summary(lm1) # DIS decreases V while stress increases V.## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: log(PhoneDuration) ~ DIS + stress + level_id + (1 | WordLabel) +
## (1 | Filename)
## Data: both_ini_V
##
## REML criterion at convergence: 838.7
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.3814 -0.4827 -0.0184 0.4708 3.4141
##
## Random effects:
## Groups Name Variance Std.Dev.
## WordLabel (Intercept) 0.250688 0.5007
## Filename (Intercept) 0.009234 0.0961
## Residual 0.170791 0.4133
## Number of obs: 546, groups: WordLabel, 159; Filename, 12
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -2.44964 0.11926 149.48769 -20.541 < 2e-16 ***
## DIS1 -0.04801 0.04865 478.58898 -0.987 0.3242
## DIS2 -0.13706 0.07670 449.44454 -1.787 0.0746 .
## DIS3 0.10055 0.10556 488.73431 0.953 0.3413
## stressstressed 0.51146 0.12356 160.13835 4.139 5.61e-05 ***
## level_id5 -0.07280 0.07902 7.95989 -0.921 0.3840
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) DIS1 DIS2 DIS3 strsss
## DIS1 -0.137
## DIS2 -0.064 0.256
## DIS3 -0.093 0.197 0.128
## strssstrssd -0.854 0.009 -0.008 0.041
## level_id5 -0.191 -0.006 -0.006 -0.029 0.000
ggplot(both_ini_V,
aes(as.factor(DIS),log(PhoneDuration), color=stress))+
geom_boxplot(outlier.shape=NA)+
xlab('Boundary Degree 0<1<2<3')+
ggtitle("V at word_initial")
lm2 = lmer(normedPhoneDur~DIS+stress+level_id+(1|WordLabel)+(1|Filename), data=both_ini_C)
lm2.2 = lmer(normedPhoneDur~DIS+stress+level_id+(1|WordLabel)+(1|Filename)+(1|PhoneLabel), data=both_ini_C)
anova(lm2,lm2.2) # The model that accounts for different phone types is better (lm2.2)## refitting model(s) with ML (instead of REML)
## Data: both_ini_C
## Models:
## lm2: normedPhoneDur ~ DIS + stress + level_id + (1 | WordLabel) + (1 | Filename)
## lm2.2: normedPhoneDur ~ DIS + stress + level_id + (1 | WordLabel) + (1 | Filename) + (1 | PhoneLabel)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## lm2 9 -394.69 -358.84 206.35 -412.69
## lm2.2 10 -480.42 -440.58 250.21 -500.42 87.731 1 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
summary(lm2.2) # DIS increases C ratio while stress decreases C ratio.## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: normedPhoneDur ~ DIS + stress + level_id + (1 | WordLabel) +
## (1 | Filename) + (1 | PhoneLabel)
## Data: both_ini_C
##
## REML criterion at convergence: -468.9
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.9949 -0.5075 -0.0767 0.5218 5.1031
##
## Random effects:
## Groups Name Variance Std.Dev.
## WordLabel (Intercept) 0.007051 0.08397
## Filename (Intercept) 0.002982 0.05461
## PhoneLabel (Intercept) 0.020076 0.14169
## Residual 0.011815 0.10870
## Number of obs: 397, groups: WordLabel, 105; Filename, 12; PhoneLabel, 7
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 0.40282 0.06285 9.88828 6.409 8.14e-05 ***
## DIS1 0.02439 0.01461 354.87006 1.669 0.096008 .
## DIS2 0.04825 0.02392 344.75368 2.017 0.044494 *
## DIS3 0.02374 0.02963 367.33844 0.801 0.423436
## stressstressed -0.11121 0.02873 91.95576 -3.870 0.000203 ***
## level_id5 0.03881 0.03897 8.59448 0.996 0.346496
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) DIS1 DIS2 DIS3 strsss
## DIS1 -0.054
## DIS2 -0.029 0.273
## DIS3 -0.062 0.183 0.131
## strssstrssd -0.371 -0.023 -0.005 0.080
## level_id5 -0.158 -0.001 -0.016 -0.005 -0.016
ggplot(both_ini_C,
aes(as.factor(DIS),normedPhoneDur, color=stress))+
geom_boxplot(outlier.shape=NA)+
xlab('Boundary Degree 0<1<2<3')+
ggtitle("C at word_initial")
For the analysis, I am interested in how proficiency level interact with stress and boundary effects. Therefore, I included interactions between predictors in the model.
lm3 = lmer(log(PhoneDuration)~DIS*stress*level_id+(1|WordLabel)+(1|Filename), data=both_ini_V)
summary(lm3) # V was lengthened by stress.## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: log(PhoneDuration) ~ DIS * stress * level_id + (1 | WordLabel) +
## (1 | Filename)
## Data: both_ini_V
##
## REML criterion at convergence: 845.6
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.4140 -0.4584 -0.0265 0.4773 3.1293
##
## Random effects:
## Groups Name Variance Std.Dev.
## WordLabel (Intercept) 0.249201 0.49920
## Filename (Intercept) 0.009455 0.09723
## Residual 0.172083 0.41483
## Number of obs: 546, groups: WordLabel, 159; Filename, 12
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -2.40613 0.13567 221.62259 -17.734 < 2e-16
## DIS1 -0.18671 0.13296 475.63407 -1.404 0.16091
## DIS2 -0.01875 0.34091 446.07475 -0.055 0.95618
## DIS3 -0.05687 0.18297 443.63131 -0.311 0.75607
## stressstressed 0.47092 0.14437 254.69596 3.262 0.00126
## level_id5 -0.12560 0.14548 80.82336 -0.863 0.39049
## DIS1:stressstressed 0.15548 0.14773 469.26282 1.052 0.29312
## DIS2:stressstressed -0.06910 0.35385 444.79964 -0.195 0.84526
## DIS3:stressstressed -0.00962 0.33319 470.57410 -0.029 0.97698
## DIS1:level_id5 0.31392 0.20148 450.29395 1.558 0.11991
## DIS2:level_id5 -0.24952 0.40566 441.96825 -0.615 0.53880
## DIS3:level_id5 0.37404 0.28396 422.29528 1.317 0.18848
## stressstressed:level_id5 0.04075 0.14618 475.15137 0.279 0.78052
## DIS1:stressstressed:level_id5 -0.36688 0.23255 451.42971 -1.578 0.11535
## DIS2:stressstressed:level_id5 0.09084 0.45245 438.90346 0.201 0.84096
## DIS3:stressstressed:level_id5 -0.11217 0.43651 460.93607 -0.257 0.79733
##
## (Intercept) ***
## DIS1
## DIS2
## DIS3
## stressstressed **
## level_id5
## DIS1:stressstressed
## DIS2:stressstressed
## DIS3:stressstressed
## DIS1:level_id5
## DIS2:level_id5
## DIS3:level_id5
## stressstressed:level_id5
## DIS1:stressstressed:level_id5
## DIS2:stressstressed:level_id5
## DIS3:stressstressed:level_id5
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation matrix not shown by default, as p = 16 > 12.
## Use print(x, correlation=TRUE) or
## vcov(x) if you need it
ggplot(both_ini_V, aes(as.factor(DIS),log(PhoneDuration), color=stress))+
geom_boxplot(outlier.shape=NA)+
xlab('Boundary Degree 0<1<2<3')+
ggtitle("V at word_initial in 1st vs. 3rd semester")+
facet_wrap(~level_id,labeller=label_both)
lm4 = lmer(normedPhoneDur~DIS*stress*level_id+(1|WordLabel)+(1|Filename), data=both_ini_C)
lm5 = lmer(normedPhoneDur~DIS*stress*level_id+(1|WordLabel)+(1|Filename)+(1|PhoneLabel), data=both_ini_C)
anova(lm4, lm5) # including PhoneLabel in random intercept made the model better## refitting model(s) with ML (instead of REML)
## Data: both_ini_C
## Models:
## lm4: normedPhoneDur ~ DIS * stress * level_id + (1 | WordLabel) + (1 | Filename)
## lm5: normedPhoneDur ~ DIS * stress * level_id + (1 | WordLabel) + (1 | Filename) + (1 | PhoneLabel)
## npar AIC BIC logLik deviance Chisq Df Pr(>Chisq)
## lm4 19 -394.77 -319.08 216.39 -432.77
## lm5 20 -485.06 -405.38 262.53 -525.06 92.288 1 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
summary(lm5) # C ratio was reduced when stressed.However, this effect disappeared when a syllable follows a pause greater than 1s. In the third semester, C ratio was reduced when the syllable follows a pause greater than 2s.## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: normedPhoneDur ~ DIS * stress * level_id + (1 | WordLabel) +
## (1 | Filename) + (1 | PhoneLabel)
## Data: both_ini_C
##
## REML criterion at convergence: -452.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.1183 -0.5232 -0.0649 0.5343 5.1312
##
## Random effects:
## Groups Name Variance Std.Dev.
## WordLabel (Intercept) 0.007108 0.08431
## Filename (Intercept) 0.003132 0.05596
## PhoneLabel (Intercept) 0.020486 0.14313
## Residual 0.011272 0.10617
## Number of obs: 397, groups: WordLabel, 105; Filename, 12; PhoneLabel, 7
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 0.405882 0.065366 11.142356 6.209 6.27e-05
## DIS1 0.028192 0.037764 348.283291 0.747 0.455855
## DIS2 -0.051921 0.072190 355.653761 -0.719 0.472473
## DIS3 0.061694 0.049972 345.420681 1.235 0.217827
## stressstressed -0.125579 0.034290 146.197962 -3.662 0.000349
## level_id5 0.061740 0.051629 24.184087 1.196 0.243352
## DIS1:stressstressed 0.012982 0.042638 350.255766 0.304 0.760946
## DIS2:stressstressed 0.134921 0.079919 349.585999 1.688 0.092262
## DIS3:stressstressed 0.164907 0.081710 310.654267 2.018 0.044431
## DIS1:level_id5 -0.014032 0.058554 335.228952 -0.240 0.810759
## DIS2:level_id5 0.048986 0.094533 331.023095 0.518 0.604674
## DIS3:level_id5 -0.189832 0.074392 326.335035 -2.552 0.011173
## stressstressed:level_id5 0.004289 0.040619 363.355094 0.106 0.915970
## DIS1:stressstressed:level_id5 -0.035543 0.067058 338.080653 -0.530 0.596438
## DIS2:stressstressed:level_id5 -0.082750 0.109125 327.045183 -0.758 0.448818
## DIS3:stressstressed:level_id5 -0.098768 0.112892 328.885315 -0.875 0.382273
##
## (Intercept) ***
## DIS1
## DIS2
## DIS3
## stressstressed ***
## level_id5
## DIS1:stressstressed
## DIS2:stressstressed .
## DIS3:stressstressed *
## DIS1:level_id5
## DIS2:level_id5
## DIS3:level_id5 *
## stressstressed:level_id5
## DIS1:stressstressed:level_id5
## DIS2:stressstressed:level_id5
## DIS3:stressstressed:level_id5
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation matrix not shown by default, as p = 16 > 12.
## Use print(x, correlation=TRUE) or
## vcov(x) if you need it
ggplot(both_ini_C,
aes(as.factor(DIS),normedPhoneDur, color=stress))+
geom_boxplot(outlier.shape=NA)+
# stat_summary(fun.data=mean_cl_boot)+
xlab('Boundary Degree 0<1<2<3')+
ggtitle("C at word_initial in 1st vs. 3rd semester")+
facet_wrap(~level_id,labeller=label_both)
lm6 = lmer(log(SyllDuration)~DIS*stress*level_id+(1|WordLabel)+(1|Filename), data=both_ini_C)
lm7 = lmer(log(SyllDuration)~DIS*stress*level_id+(1|WordLabel)+(1|Filename)+(1|PhoneLabel), data=both_ini_C)
summary(lm7)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: log(SyllDuration) ~ DIS * stress * level_id + (1 | WordLabel) +
## (1 | Filename) + (1 | PhoneLabel)
## Data: both_ini_C
##
## REML criterion at convergence: 544.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.8556 -0.4768 -0.0356 0.4323 7.4015
##
## Random effects:
## Groups Name Variance Std.Dev.
## WordLabel (Intercept) 0.1129918 0.3361
## Filename (Intercept) 0.0005854 0.0242
## PhoneLabel (Intercept) 0.0254051 0.1594
## Residual 0.1650928 0.4063
## Number of obs: 397, groups: WordLabel, 105; Filename, 12; PhoneLabel, 7
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -1.72846 0.13390 45.87145 -12.908 < 2e-16
## DIS1 0.05621 0.14031 341.69393 0.401 0.68896
## DIS2 -0.07402 0.27353 361.61660 -0.271 0.78685
## DIS3 -0.03739 0.18695 349.02945 -0.200 0.84162
## stressstressed 0.39690 0.13072 165.50026 3.036 0.00278
## level_id5 -0.08678 0.13715 173.30538 -0.633 0.52775
## DIS1:stressstressed 0.15403 0.15876 346.19042 0.970 0.33263
## DIS2:stressstressed 0.29959 0.30199 357.34679 0.992 0.32184
## DIS3:stressstressed 0.07718 0.30901 320.20061 0.250 0.80292
## DIS1:level_id5 0.02667 0.21964 339.17959 0.121 0.90343
## DIS2:level_id5 -0.20240 0.35914 339.94108 -0.564 0.57342
## DIS3:level_id5 0.13254 0.28220 334.94864 0.470 0.63890
## stressstressed:level_id5 0.10467 0.15464 372.52970 0.677 0.49893
## DIS1:stressstressed:level_id5 -0.14307 0.25322 345.02124 -0.565 0.57243
## DIS2:stressstressed:level_id5 0.02894 0.41494 338.23473 0.070 0.94444
## DIS3:stressstressed:level_id5 -0.07471 0.43041 336.42580 -0.174 0.86230
##
## (Intercept) ***
## DIS1
## DIS2
## DIS3
## stressstressed **
## level_id5
## DIS1:stressstressed
## DIS2:stressstressed
## DIS3:stressstressed
## DIS1:level_id5
## DIS2:level_id5
## DIS3:level_id5
## stressstressed:level_id5
## DIS1:stressstressed:level_id5
## DIS2:stressstressed:level_id5
## DIS3:stressstressed:level_id5
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation matrix not shown by default, as p = 16 > 12.
## Use print(x, correlation=TRUE) or
## vcov(x) if you need it
ggplot(both_ini_C,
aes(as.factor(DIS),log(SyllDuration), color=stress))+
geom_boxplot(outlier.shape=NA)+
# stat_summary(fun.data=mean_cl_boot)+
xlab('Boundary Degree 0<1<2<3')+
ggtitle("C at word_initial in 1st vs. 3rd semester")+
facet_wrap(~level_id,labeller=label_both)