This week’s goals
- Fully understand my code for my 1st exploratory question
- Continue brainstorming ideas for my exploratory analysis, aim to at least think of 2 more ideas even if I can’t finish all the code by the end of this week
Brainstoriming ideas for exploratory analysis
See if there is any other variable related to sleep that produces a significant correlation with differential bias change. The authors explored the correlations between each type of sleep stage recorded in their data with differential bias change and none of them were statistically significant
I wonder if there are differences between men and women in terms of their implicit bias levels following TMR or counter stereotype training only (averaged across gender and race implicit bias). This idea is still a work in progress, there are lots of time frames, and different types of implicit bias measures that I could use, so I’ll need to think of some sort narrower direction and a rationale to go along with it before I start coding.
Are there differences in the effectiveness of TMR and counter stereotype training only depending on experiment compensation. This is my back up question if I can’t figure out some of the other questions.
loading libraries
I’m going to first load my libraries for coding later
library(tidyverse) #for dyplr and ggplot to conduct data wrangling and visualization## -- Attaching packages --------------------------------------- tidyverse 1.3.1 --## v ggplot2 3.3.5 v purrr 0.3.4
## v tibble 3.1.2 v dplyr 1.0.6
## v tidyr 1.1.3 v stringr 1.4.0
## v readr 1.4.0 v forcats 0.5.1## -- Conflicts ------------------------------------------ tidyverse_conflicts() --
## x dplyr::filter() masks stats::filter()
## x dplyr::lag() masks stats::lag()library(gt) #for creating highly customisable tables
library(janitor) #to clean names##
## Attaching package: 'janitor'## The following objects are masked from 'package:stats':
##
## chisq.test, fisher.testlibrary(rstatix) #for inferrential statistics##
## Attaching package: 'rstatix'## The following object is masked from 'package:janitor':
##
## make_clean_names## The following object is masked from 'package:stats':
##
## filterlibrary(jmv) #for inferrential statistics
library(ggpubr) #for t tests and other inferrential statistics
library(ggeasy) #for easy ggplot functions
library(corrplot) #for correlation matrixes## corrplot 0.89 loadedQuestion 2: Do other measures of sleep quality correlate with differential bias change?
In part 2 of the verification report, our group reproduced this graph:
This graph shows the insignificant correlation between SWS x REM sleep duration (min) and differential bias change (the difference between the cued and uncued conditions regarding their respective change in implicit bias levels across time). The authors then explored whether time spent in other stages of sleep were correlated to differential bias change. They didn’t find a significant correlation between the time spent in any of the other sleep stages (non-rem 1, non-rem 2, SWS (slow wave sleep) individually, REM sleep individually) and differential bias change. They also didn’t find a significant correlation between the length of time the cue was presented during sleep and differential bias change (though the p value for this correlation was the closest to being significant compared to the other correlations).
When revising this graph again, I was curious as to why the authors measured time in slow wave sleep (SWS) multiplied by time in REM sleep because this didn’t really make much sense to me as what it actually measured. I first thought perhaps they did that because SWS and REM are correlated moreso than the other sleep stages are. So to check, I used the cor function from the corrplot package to create a correlation matrix between time spent in all of the sleep stages. select allows me to select values from those variables to create the matrix from:
cleandata <- read.csv("cleandata.csv") #reading in the cleandata frame saved from earlier
q2_corr_matrix <- cleandata %>% #I decided to use cleandata because I didn't want to change the exclusion criteria from what the authors originally decided on
select(total_sleep, wake_amount, nrem1_amount, nrem2_amount, sws_amount, rem_amount) %>%
cor()
corrplot(q2_corr_matrix)
So it turns out that this isn’t true and the time spent in each of the sleep stages didn’t really correlate with any other ones. So I had to refer to the original paper (Hu et al., 2015) that ours replicated to figure out that SWS x REM was used used as it was a measure of sleep quality with a higher value indicating greater sleep quality.
This is intriguing because other measures of sleep quality may correlate with differential bias change. One other measure of sleep quality is the proportion of time spent in REM sleep. It seems that healthy sleep is determined by REM sleep approx. accounting for 20-25% of the total time asleep in adults (Altevogt & Colten, 2006). Another measure of sleep quality is the total amount of time asleep compared to time awake.
This brings me to my second question, whether other measures of sleep quality are correlated with differential bias change.
obtaining relevant variables
First I am selecting variables that will be useful for this question using select from the ‘cleandata’ dataframe, as I am still following the exlucsion criteria of the authors.These variables are the differential bias change value (diff_biaschange), time spent sleeping, awake, in REM and time in SWS x REM.
I used rename to renmame the sws x rem variable so that it’s easier to work with.
Then I used mutate to create three new columns/variables of data. The first one is ‘rem_pct’ which is the total amount of time the participant spends in rem compared to their total time asleep. The second one is ‘sleep_and_wake’ which is the total time the participants spent awake and alseep. This value is used to create the third variable ‘asleep_pct’ which is the percentage of time that the participant spends asleep compared to awake.
q2 <- cleandata %>%
select(diff_biaschange, diff_biaschange_cued, diff_biaschange_uncued, total_sleep, wake_amount, rem_amount, sw_sx_rem, cue_minutes) %>%
rename(sws_x_rem = sw_sx_rem) %>%
mutate(rem_pct = rem_amount/total_sleep * 100,
sleep_and_wake = total_sleep + wake_amount,
asleep_pct = total_sleep/sleep_and_wake * 100)
print(q2)## diff_biaschange diff_biaschange_cued diff_biaschange_uncued total_sleep
## 1 0.44490584 0.371668148 -0.07323769 65
## 2 -1.01429471 -0.359624732 0.65466998 66
## 3 -1.00530440 -0.192821041 0.81248335 80
## 4 -0.49923043 -0.570273967 -0.07104354 62
## 5 0.67261133 0.590693579 -0.08191775 51
## 6 -0.03539601 -0.250479588 -0.21508358 81
## 7 -1.47683672 0.300986111 1.77782283 81
## 8 -0.15781907 0.425109806 0.58292888 67
## 9 -0.53136657 0.309977544 0.84134411 79
## 10 0.22017019 0.472346888 0.25217670 71
## 11 0.80800876 0.636426002 -0.17158276 68
## 12 0.71564723 0.172082405 -0.54356483 85
## 13 0.69400342 0.531805761 -0.16219766 74
## 14 0.36763005 0.725511445 0.35788139 73
## 15 -1.11173748 -0.793593406 0.31814407 79
## 16 -0.82626125 -0.920808186 -0.09454693 37
## 17 -0.29533509 -0.037791997 0.25754309 52
## 18 1.26682255 1.019294435 -0.24752811 62
## 19 0.25179469 -0.002406687 -0.25420138 88
## 20 -0.23329644 -0.362653222 -0.12935678 70
## 21 0.04315263 0.055971988 0.01281936 77
## 22 0.17642051 0.507566544 0.33114603 74
## 23 0.02712447 1.220839300 1.19371483 77
## 24 -0.53384387 -0.357015106 0.17682877 77
## 25 -0.13485088 -0.183065550 -0.04821467 84
## 26 0.34427330 0.564860845 0.22058754 79
## 27 -0.31348070 -0.127572774 0.18590793 65
## 28 0.20937037 -0.140763994 -0.35013437 61
## 29 -0.78933933 -0.093405722 0.69593361 75
## 30 0.28747433 0.282019388 -0.00545494 80
## 31 -0.01955842 -0.148474613 -0.12891620 84
## wake_amount rem_amount sws_x_rem cue_minutes rem_pct sleep_and_wake
## 1 25 23 276 9.5 35.384615 90
## 2 24 0 0 12.0 0.000000 90
## 3 10 17 408 15.5 21.250000 90
## 4 28 17 408 16.0 27.419355 90
## 5 39 2 32 15.0 3.921569 90
## 6 9 18 648 25.0 22.222222 90
## 7 9 9 333 23.0 11.111111 90
## 8 23 9 36 3.0 13.432836 90
## 9 11 24 552 18.5 30.379747 90
## 10 19 11 275 19.0 15.492958 90
## 11 22 10 70 12.5 14.705882 90
## 12 5 16 496 28.0 18.823529 90
## 13 16 0 0 32.5 0.000000 90
## 14 17 17 476 28.0 23.287671 90
## 15 11 11 363 36.0 13.924051 90
## 16 53 0 0 14.0 0.000000 90
## 17 38 9 198 19.0 17.307692 90
## 18 28 0 0 2.5 0.000000 90
## 19 2 9 216 26.0 10.227273 90
## 20 20 16 240 18.0 22.857143 90
## 21 13 10 450 44.0 12.987013 90
## 22 16 18 684 29.5 24.324324 90
## 23 13 22 836 37.5 28.571429 90
## 24 13 23 506 28.5 29.870130 90
## 25 6 0 0 40.0 0.000000 90
## 26 11 19 418 3.5 24.050633 90
## 27 25 24 480 27.0 36.923077 90
## 28 29 5 50 11.0 8.196721 90
## 29 15 0 0 33.5 0.000000 90
## 30 10 12 336 17.5 15.000000 90
## 31 6 14 224 23.5 16.666667 90
## asleep_pct
## 1 72.22222
## 2 73.33333
## 3 88.88889
## 4 68.88889
## 5 56.66667
## 6 90.00000
## 7 90.00000
## 8 74.44444
## 9 87.77778
## 10 78.88889
## 11 75.55556
## 12 94.44444
## 13 82.22222
## 14 81.11111
## 15 87.77778
## 16 41.11111
## 17 57.77778
## 18 68.88889
## 19 97.77778
## 20 77.77778
## 21 85.55556
## 22 82.22222
## 23 85.55556
## 24 85.55556
## 25 93.33333
## 26 87.77778
## 27 72.22222
## 28 67.77778
## 29 83.33333
## 30 88.88889
## 31 93.33333identifying outliers
Then I’m using the identify_outliers from the rstatix package to see if anyone’s ’asleep_pct` is an outlier.
q2_outliers <- q2 %>%
identify_outliers(asleep_pct)
print(q2_outliers)## diff_biaschange diff_biaschange_cued diff_biaschange_uncued total_sleep
## 1 -0.8262613 -0.9208082 -0.09454693 37
## wake_amount rem_amount sws_x_rem cue_minutes rem_pct sleep_and_wake
## 1 53 0 0 14 0 90
## asleep_pct is.outlier is.extreme
## 1 41.11111 TRUE FALSEThere are no extreme outliers, so I will not be filtering this dataset any further.
data visualisation
Now to create a ggplot to visualise the correlation between quality of sleep defined as percentage of time spent asleep, and differential bias change:
I’m using geom_point to create a scatterplot, and adding another layer geom_smooth to add the line of best fit, specifing the line to be linear with method = lm, and removing the confidence interval shading with se = F.
q2.1_pic <- ggplot(q2, aes(asleep_pct, diff_biaschange))+
geom_point()+
geom_smooth(method = lm, se = F)
plot(q2.1_pic)## `geom_smooth()` using formula 'y ~ x'
another <- ggplot(q2, aes(asleep_pct, diff_biaschange_cued))+
geom_point()+
geom_smooth(method = lm, se = F)
plot(another)## `geom_smooth()` using formula 'y ~ x'
There doesn’t seem to be a correlation between these two variables. Before moving onto inferrential statistics, lets check how many people fall into the 20-25% of REM sleep using the same ggplot functions as the previous graph:
q2_pic <- ggplot(q2, aes(rem_pct, diff_biaschange))+
geom_point()+
geom_smooth(method = lm, se = F)
plot(q2_pic)## `geom_smooth()` using formula 'y ~ x'
It doesn’t look like many people do.
Originally I was going to use filter the data from the q2 data frame by excluding all participants who did not have good quality sleep by filtering out participants whose rem_pct was beyond the 20-25% range as previously stated and then using that sample of participants to investigate the correlation between SWSxREM and differential bias change. However as we can see from the graph above, only 6 participants meet this criteria, and therefore wouldn’t be a large enough sample to conduct my intended analysis.
Therefore I will be only using statistical analysis to look at the correlation between asleep_pct and differential bias change.
inferential statistics
I’m using the function lm from the package stats to carry out a regression between asleep_pct and diff_bias change as seen below
q2_stats <- lm(asleep_pct ~ diff_biaschange, data = q2)
summary(q2_stats)##
## Call:
## lm(formula = asleep_pct ~ diff_biaschange, data = q2)
##
## Residuals:
## Min 1Q Median 3Q Max
## -39.047 -7.058 2.969 8.470 18.261
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 79.6662 2.3039 34.578 <2e-16 ***
## diff_biaschange -0.5955 3.6589 -0.163 0.872
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 12.73 on 29 degrees of freedom
## Multiple R-squared: 0.0009124, Adjusted R-squared: -0.03354
## F-statistic: 0.02648 on 1 and 29 DF, p-value: 0.8719As expected after looking at my graph,
The correlation between quality of sleep (proportion of time spent asleep comapred to awake) and differential bias change is not significant (F = 0.02648, df = 29, p-value = 0.8719, R^2 = -0.03354)
conclusions Other measures of quality of sleep (proportion of time spent asleep compared to awake) are not correlated with differential bias change just like SWS x REM (mins).
thoughts The p-value obtained is quite a large so instead of doing further exploratory analysis on this measure of quality of sleep, I wanted to follow up on the correlation between minutes of the cue presented and differential bias change since it did have the lowest p-value out of all the correlations explored by the authors.
follow up - replicating exploratory analysis from the authors: the correlation between cue_minutes and differential bias change
visualising the data
Because the q2 dataframe I made earlier has the relevant x and y vairables (cue_minutes and diff_biaschange) I am creating a ggplot similar to all previous graphs from that data frame.
q2_cue <- ggplot(q2, aes(cue_minutes, diff_biaschange))+
geom_point()+
geom_smooth(method = lm, se = F)
plot(q2_cue)## `geom_smooth()` using formula 'y ~ x'
It looks like there is a negative correlation here, so I’m going to use ‘cor’ just like before to see if I can reproduce the values from the paper
q2_cue_stats <- lm(cue_minutes ~ diff_biaschange, data = q2)
summary(q2_cue_stats)##
## Call:
## lm(formula = cue_minutes ~ diff_biaschange, data = q2)
##
## Residuals:
## Min 1Q Median 3Q Max
## -18.810 -7.908 -2.911 8.277 22.815
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 21.319 1.959 10.882 9.38e-12 ***
## diff_biaschange -3.109 3.111 -0.999 0.326
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 10.82 on 29 degrees of freedom
## Multiple R-squared: 0.03328, Adjusted R-squared: -5.004e-05
## F-statistic: 0.9985 on 1 and 29 DF, p-value: 0.3259conclusions
Total number of minutes the cue is played for during sleep is not correlated with differential bias change (F = 0.9985, df = 29, p-value = 0.326 , R^2 = 0.03328. Checking to results from the paper, these results match.
following on
But I want to take things one step further and apply the same logic as previous analysis, to look at the percentage of time that the cue was presented compared to the total amount of time asleep, rather than just looking at the total time the cue was presented. The reason for this is because, the rationale behind TMR (targeted memory reactivation) is that sleep helps strengthen the memory of counter stereotype training by played a cue associated to that training during sleep. Since the participants were asleep over large range of times, and had the cue presented to them across a range of times, perhaps looking at the ratio between cue played (mins) and time asleep (total_sleep) would result in a stronger correlation to differntial bias change (diff_biaschange).
obtaining my variables
First I’m creating a subset dataframe from q2 where I create a new variable uising mutate which has the equation to calculate the percentage of time where the cue was played whilst the participant was asleepl.
q2_cueextra <- q2 %>%
mutate(cue_pct = cue_minutes/total_sleep*100)
print(q2_cueextra)## diff_biaschange diff_biaschange_cued diff_biaschange_uncued total_sleep
## 1 0.44490584 0.371668148 -0.07323769 65
## 2 -1.01429471 -0.359624732 0.65466998 66
## 3 -1.00530440 -0.192821041 0.81248335 80
## 4 -0.49923043 -0.570273967 -0.07104354 62
## 5 0.67261133 0.590693579 -0.08191775 51
## 6 -0.03539601 -0.250479588 -0.21508358 81
## 7 -1.47683672 0.300986111 1.77782283 81
## 8 -0.15781907 0.425109806 0.58292888 67
## 9 -0.53136657 0.309977544 0.84134411 79
## 10 0.22017019 0.472346888 0.25217670 71
## 11 0.80800876 0.636426002 -0.17158276 68
## 12 0.71564723 0.172082405 -0.54356483 85
## 13 0.69400342 0.531805761 -0.16219766 74
## 14 0.36763005 0.725511445 0.35788139 73
## 15 -1.11173748 -0.793593406 0.31814407 79
## 16 -0.82626125 -0.920808186 -0.09454693 37
## 17 -0.29533509 -0.037791997 0.25754309 52
## 18 1.26682255 1.019294435 -0.24752811 62
## 19 0.25179469 -0.002406687 -0.25420138 88
## 20 -0.23329644 -0.362653222 -0.12935678 70
## 21 0.04315263 0.055971988 0.01281936 77
## 22 0.17642051 0.507566544 0.33114603 74
## 23 0.02712447 1.220839300 1.19371483 77
## 24 -0.53384387 -0.357015106 0.17682877 77
## 25 -0.13485088 -0.183065550 -0.04821467 84
## 26 0.34427330 0.564860845 0.22058754 79
## 27 -0.31348070 -0.127572774 0.18590793 65
## 28 0.20937037 -0.140763994 -0.35013437 61
## 29 -0.78933933 -0.093405722 0.69593361 75
## 30 0.28747433 0.282019388 -0.00545494 80
## 31 -0.01955842 -0.148474613 -0.12891620 84
## wake_amount rem_amount sws_x_rem cue_minutes rem_pct sleep_and_wake
## 1 25 23 276 9.5 35.384615 90
## 2 24 0 0 12.0 0.000000 90
## 3 10 17 408 15.5 21.250000 90
## 4 28 17 408 16.0 27.419355 90
## 5 39 2 32 15.0 3.921569 90
## 6 9 18 648 25.0 22.222222 90
## 7 9 9 333 23.0 11.111111 90
## 8 23 9 36 3.0 13.432836 90
## 9 11 24 552 18.5 30.379747 90
## 10 19 11 275 19.0 15.492958 90
## 11 22 10 70 12.5 14.705882 90
## 12 5 16 496 28.0 18.823529 90
## 13 16 0 0 32.5 0.000000 90
## 14 17 17 476 28.0 23.287671 90
## 15 11 11 363 36.0 13.924051 90
## 16 53 0 0 14.0 0.000000 90
## 17 38 9 198 19.0 17.307692 90
## 18 28 0 0 2.5 0.000000 90
## 19 2 9 216 26.0 10.227273 90
## 20 20 16 240 18.0 22.857143 90
## 21 13 10 450 44.0 12.987013 90
## 22 16 18 684 29.5 24.324324 90
## 23 13 22 836 37.5 28.571429 90
## 24 13 23 506 28.5 29.870130 90
## 25 6 0 0 40.0 0.000000 90
## 26 11 19 418 3.5 24.050633 90
## 27 25 24 480 27.0 36.923077 90
## 28 29 5 50 11.0 8.196721 90
## 29 15 0 0 33.5 0.000000 90
## 30 10 12 336 17.5 15.000000 90
## 31 6 14 224 23.5 16.666667 90
## asleep_pct cue_pct
## 1 72.22222 14.615385
## 2 73.33333 18.181818
## 3 88.88889 19.375000
## 4 68.88889 25.806452
## 5 56.66667 29.411765
## 6 90.00000 30.864198
## 7 90.00000 28.395062
## 8 74.44444 4.477612
## 9 87.77778 23.417722
## 10 78.88889 26.760563
## 11 75.55556 18.382353
## 12 94.44444 32.941176
## 13 82.22222 43.918919
## 14 81.11111 38.356164
## 15 87.77778 45.569620
## 16 41.11111 37.837838
## 17 57.77778 36.538462
## 18 68.88889 4.032258
## 19 97.77778 29.545455
## 20 77.77778 25.714286
## 21 85.55556 57.142857
## 22 82.22222 39.864865
## 23 85.55556 48.701299
## 24 85.55556 37.012987
## 25 93.33333 47.619048
## 26 87.77778 4.430380
## 27 72.22222 41.538462
## 28 67.77778 18.032787
## 29 83.33333 44.666667
## 30 88.88889 21.875000
## 31 93.33333 27.976190Then I’m using the same ggplot as before to visualse the correlation
q2_cueextra_pic <- ggplot(q2_cueextra, aes(cue_pct, diff_biaschange))+
geom_point()+
geom_smooth(method = lm, se = F)
plot(q2_cueextra_pic)## `geom_smooth()` using formula 'y ~ x'
It seems that it is more sloped than the cue_minutes x diff_biaschange graph. To test if the correlation is significant I used the same cor function as previously:
q2_cueextra <- lm(cue_pct ~ diff_biaschange, data = q2_cueextra)
summary(q2_cueextra)##
## Call:
## lm(formula = cue_pct ~ diff_biaschange, data = q2_cueextra)
##
## Residuals:
## Min 1Q Median 3Q Max
## -25.672 -8.277 1.298 10.678 27.951
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 29.398 2.411 12.192 6.16e-13 ***
## diff_biaschange -4.766 3.829 -1.245 0.223
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 13.32 on 29 degrees of freedom
## Multiple R-squared: 0.0507, Adjusted R-squared: 0.01797
## F-statistic: 1.549 on 1 and 29 DF, p-value: 0.2233Ok the p-value is lower but it still isn’t significant.
conclusions
The proportion of time asleep that the cue is played for is not correlated with differential bias change (F = 1.549, df = 29, p-value = 0.2233 , R^2 = 0.0507. Checking to results from the paper, these results match.
Challenges and Successess
Wow this week was tough. I had some troubles thinking of cool exploratory analysis questions that had a rationale to back them up. Hopefully what I’ve got is good enough.
I’m not so sure that my logic follows through for this question, especially since it’s a little different from my first question in the sense that it has a sort of follow up section with further analysis, and I have also omitted statistical analysis in part of the code (though I did provide a reason for it)
I’m proud of myself for working on a question even though I originally wanted to explore 2 more questions, it has been a busy and tough week for me.
Questions
I’m would like to know if the logic for this exploratory question works/makes sense as I do tend to get lost in what I’m doing sometimes if I spend a lot of time on it so a fresh pair of eyes would be great :)
When reporting the R squared value, which one is the appropriate one to use - the multiple R squared value or the adjusted R squared value?
Next week
Tidy up code by revising how to make my own function so I don’t have to type very similar code for very similar graphs 4 times, and adding aesthetics to my graphs
Make a start on my final exploratory analysis question and finish the code/my first attempt in time for the qna session