SOL-analysis
Kyle MacDonald
June 9, 2015
Data processing, plotting, and analysis script
rm(list=ls())
knitr::opts_chunk$set(warning=FALSE, message=FALSE, sanitize = T,
fig.height=8, fig.width=10, echo=F, cache = T)## [1] "loading library"Read Data
Demographics
Eye movement data
Stimuli information
Filter dataset
Filter out participants that should not go into analyses based on exclusionary criteria: a) age, b) didn’t know signs in the task, c) not enough ASL exposure.
Remove unknown signs
Create a clean target image variable.
Add unknown signs variable. Taken from sol_demo data frame.
Now we can filter the iChart, removing the “unknown” signs.
Get total number of trials removed because of known signs
## [1] 22Remove prescreened out trials
Remove participants for having too few trials
We define too few trials as less than or equal to 25% of the total number of trials in the task.
## Source: local data frame [2 x 2]
##
## exclude_few_trials n()
## (chr) (int)
## 1 exclude 5
## 2 include 45Create final exclusions table
Get median split by age
## median(Months) max(Months) min(Months) sd(Months) n()
## 1 27 53 16 9.140259 29## Source: local data frame [4 x 3]
## Groups: Sex [?]
##
## Sex age_group count
## (chr) (chr) (int)
## 1 F < 27 Months 8
## 2 F >= 27 Months 9
## 3 M < 27 Months 6
## 4 M >= 27 Months 6## Source: local data frame [3 x 5]
##
## age_group n_distinct(Sub.Num) mean(Months) min(Months) max(Months)
## (chr) (int) (dbl) (int) (int)
## 1 < 27 Months 14 20.71308 16 26
## 2 >= 27 Months 15 36.18715 27 53
## 3 Adults 16 430.57634 246 695Process iChart
First, we need to process the data, keeping only those trials on which the child was looking at the signer at F0.
- C: Center
- D: Distractor
- T: Target
- A: Away
includeOffCenter == FALSE -> only include trials child was looking at center at F0
includeOffCenter == TRUE -> include trials child was looking at center, target, or distractor at F0
## Source: local data frame [4 x 3]
## Groups: "0" [?]
##
## "0" Response Trials
## (chr) (chr) (int)
## 1 0 A 22
## 2 0 C 1019
## 3 0 D 35
## 4 0 T 43## Source: local data frame [2 x 3]
## Groups: "0" [?]
##
## "0" Response Trials
## (chr) (chr) (int)
## 1 0 A 100
## 2 0 D 1019Flag C-T and C-D Trials
Datawiz does not tell us which shifts land on a target vs. a disctractor. So we need to use a function that flags each trial as one of the following:
- C_T: center to target
- C_D: center to distractor
- C-C: center to center (child leaves the signer, goes away, and comes back to signer)
- no_shift
- off_signer
## Source: local data frame [5 x 2]
##
## trial_types Trials
## (fctr) (int)
## 1 C_C 1
## 2 C_D 141
## 3 C_T 845
## 4 no_shift 32
## 5 off_signer 100Next, we compute statistics over long window 0-5000 ms. This will allow us to see a distribution of RTs, which we will use to determine our analysis window.
First, for adults.
## [1] "### Trials left ###"
## [1] 500
## [1] 450
## [1] 400
## [1] 350
## [1] 300
## [1] 250
## [1] 200
## [1] 150
## [1] 100
## [1] 50
## [1] 0
Now do the same for kids.
## [1] "### Trials left ###"
## [1] 550
## [1] 500
## [1] 450
## [1] 400
## [1] 350
## [1] 300
## [1] 250
## [1] 200
## [1] 150
## [1] 100
## [1] 50
## [1] 0
Compute statistics over analysis window: 0-2700ms. We use 2700 ms because it is 500 ms longer than the end of our analysis window (2200ms). This allows us to include trials in which the participant to initiates and completes a shift at the very end of the analysis window.
## [1] "### Trials left ###"
## [1] 1100
## [1] 1050
## [1] 1000
## [1] 950
## [1] 900
## [1] 850
## [1] 800
## [1] 750
## [1] 700
## [1] 650
## [1] 600
## [1] 550
## [1] 500
## [1] 450
## [1] 400
## [1] 350
## [1] 300
## [1] 250
## [1] 200
## [1] 150
## [1] 100
## [1] 50
## [1] 0## [1] "### Trials left ###"
## [1] 550
## [1] 500
## [1] 450
## [1] 400
## [1] 350
## [1] 300
## [1] 250
## [1] 200
## [1] 150
## [1] 100
## [1] 50
## [1] 0## [1] "### Trials left ###"
## [1] 500
## [1] 450
## [1] 400
## [1] 350
## [1] 300
## [1] 250
## [1] 200
## [1] 150
## [1] 100
## [1] 50
## [1] 0Reject trials with really long RTs and with long gaps. Gaps are defined as a sequence of frames when the child is not looking at either picture or at the signer.
Mean Accuracy and RT for each participant
Mean Acc and RT for each participant by condition
Statistics
Get mean accuracy and rt for each participant
Some munging to get data frame for analysis. Variables needed for each subject:
- Mean acc
- Mean rt
- Signs produced
- Age
- Age condition
First shift accuracy analysis
First we need to categorize each trial type based on the analysis window.
Then we add this information to the summary data frame.
Median RT
Filter kids for RT analysis
Correlations
## mean_accuracy C_T_prop mean_correct_rt median_ct_rt
## mean_accuracy 1.00 0.58 -0.54 -0.49
## C_T_prop 0.58 1.00 -0.02 -0.14
## mean_correct_rt -0.54 -0.02 1.00 0.94
## median_ct_rt -0.49 -0.14 0.94 1.00
## signs_produced 0.44 0.31 -0.41 -0.46
## Months 0.62 0.38 -0.32 -0.36
## signs_produced Months
## mean_accuracy 0.44 0.62
## C_T_prop 0.31 0.38
## mean_correct_rt -0.41 -0.32
## median_ct_rt -0.46 -0.36
## signs_produced 1.00 0.76
## Months 0.76 1.00
##
## n
## mean_accuracy C_T_prop mean_correct_rt median_ct_rt
## mean_accuracy 29 29 29 29
## C_T_prop 29 29 29 29
## mean_correct_rt 29 29 29 29
## median_ct_rt 29 29 29 29
## signs_produced 28 28 28 28
## Months 29 29 29 29
## signs_produced Months
## mean_accuracy 28 29
## C_T_prop 28 29
## mean_correct_rt 28 29
## median_ct_rt 28 29
## signs_produced 28 28
## Months 28 29
##
## P
## mean_accuracy C_T_prop mean_correct_rt median_ct_rt
## mean_accuracy 0.0010 0.0027 0.0067
## C_T_prop 0.0010 0.9112 0.4565
## mean_correct_rt 0.0027 0.9112 0.0000
## median_ct_rt 0.0067 0.4565 0.0000
## signs_produced 0.0205 0.1122 0.0298 0.0132
## Months 0.0003 0.0436 0.0922 0.0553
## signs_produced Months
## mean_accuracy 0.0205 0.0003
## C_T_prop 0.1122 0.0436
## mean_correct_rt 0.0298 0.0922
## median_ct_rt 0.0132 0.0553
## signs_produced 0.0000
## Months 0.0000
Just RT.
## mean_correct_rt median_ct_rt signs_produced Months
## mean_correct_rt 1.00 0.92 -0.61 -0.47
## median_ct_rt 0.92 1.00 -0.65 -0.49
## signs_produced -0.61 -0.65 1.00 0.74
## Months -0.47 -0.49 0.74 1.00
##
## n
## mean_correct_rt median_ct_rt signs_produced Months
## mean_correct_rt 24 24 23 24
## median_ct_rt 24 24 23 24
## signs_produced 23 23 23 23
## Months 24 24 23 24
##
## P
## mean_correct_rt median_ct_rt signs_produced Months
## mean_correct_rt 0.0000 0.0021 0.0204
## median_ct_rt 0.0000 0.0009 0.0150
## signs_produced 0.0021 0.0009 0.0000
## Months 0.0204 0.0150 0.0000
T.tests
##
## Two Sample t-test
##
## data: mean_correct_rt by age_group
## t = 2.2137, df = 22, p-value = 0.03752
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 12.11964 371.55653
## sample estimates:
## mean in group < 27 Months mean in group >= 27 Months
## 1429.099 1237.261##
## Two Sample t-test
##
## data: mean_accuracy by age_group
## t = -2.8347, df = 27, p-value = 0.008581
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.12566730 -0.02013377
## sample estimates:
## mean in group < 27 Months mean in group >= 27 Months
## 0.6040133 0.6769139##
## Two Sample t-test
##
## data: mean_correct_rt by age_group_collapsed
## t = -8.8497, df = 43, p-value = 0.00000000003075
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -720.4198 -453.0145
## sample estimates:
## mean in group Adults mean in group Kids
## 716.6112 1303.3284##
## Two Sample t-test
##
## data: mean_accuracy by hearing_status_participant
## t = 1.1101, df = 27, p-value = 0.2767
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.02710879 0.09101862
## sample estimates:
## mean in group deaf mean in group hearing
## 0.6560451 0.6240902##
## Two Sample t-test
##
## data: C_T_prop by hearing_status_participant
## t = 1.7148, df = 27, p-value = 0.09785
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.01990326 0.22240326
## sample estimates:
## mean in group deaf mean in group hearing
## 0.76125 0.66000##
## Two Sample t-test
##
## data: mean_correct_rt by hearing_status_participant
## t = 0.65197, df = 27, p-value = 0.5199
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -120.1615 232.0884
## sample estimates:
## mean in group deaf mean in group hearing
## 1328.415 1272.452Get confidence intervals for graph values
PP for each participant
Some munging to format data for plotting
dplyr::summarise graph values for all participants.
PP for children (younger, older) and adults.
Summary bar plots for all participants
Now we make the summary bar plots.
Tanenhaus plot: all types of looking
First, we munge the raw iChart data.
Grab just the eye movement data and group information
Convert to long format
Summarize for each participant - get proportion looking at each time slice
Get means and CIs for proportion looking at each time slice across particpants
Now we make Tanenhaus style plot.
Window accuracy analysis
Here I want to see what happens to window accuracy when we don’t penalize kids for looking at the signer.
Grab just the eye movement data and group information
Convert to long format
Get means and CIs for each participant.
Get means and CIS collapsing across participants
Now plot.
CODAs vs. Deaf Analysis
Summary data plots for CODAs vs. Deaf children.
T.tests deaf vs. codas
##
## Two Sample t-test
##
## data: mean_accuracy by hearing_status_participant
## t = 1.1101, df = 27, p-value = 0.2767
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.02710879 0.09101862
## sample estimates:
## mean in group deaf mean in group hearing
## 0.6560451 0.6240902##
## Two Sample t-test
##
## data: C_T_prop by hearing_status_participant
## t = 1.7148, df = 27, p-value = 0.09785
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.01990326 0.22240326
## sample estimates:
## mean in group deaf mean in group hearing
## 0.76125 0.66000##
## Two Sample t-test
##
## data: mean_correct_rt by hearing_status_participant
## t = 0.65197, df = 27, p-value = 0.5199
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -120.1615 232.0884
## sample estimates:
## mean in group deaf mean in group hearing
## 1328.415 1272.452Effect size deaf vs. codas
PP for CODAs vs. Deaf
PP for CODAs vs. Deaf Signer removed
Get means and CIs for each participant for each time slice. Using two different Accuracy computations:
- Including looks to the signer
- Not including looks to the signer
Get means and CIs for each group
Now plot the two accuracy measures against each other.
Summary plots comparing accuracy measures: CODAs vs. Deaf
##
## Two Sample t-test
##
## data: m_acc by hearing_status_participant
## t = 1.2432, df = 27, p-value = 0.2245
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.03895116 0.15870804
## sample estimates:
## mean in group deaf mean in group hearing
## 0.8134356 0.7535572Tanenhaus plot codas vs. deaf
Individual Participants Analysis
Item-level Analysis
PP by item
Read in graph values.
Munge data for plotting
Plot by item data
Plot all items for both stimulus sets
Plot RT distribution for each stimulus set.
Add sign length information to iChart.
Get only good RTs: Center to Target shifts, within the analysis window.
Plot RT distribution for each stimulus set and each sign.
Plot difference between participants’ mean RT by item and offset of that sign.
Scatter plots of individual differences
