Setup & Data Import

library(readxl)
library(tidyverse)
library(supernova)
library(mosaic)
library(ggplot2)
library(WRS2)
library(knitr)

participant_info_midterm <- read_excel("~/Downloads/participant_info_midterm.xlsx")
sleep_data_midterm <- read_excel("~/Downloads/sleep_data_midterm.xlsx")
kable(head(participant_info_midterm[, 1:4]), caption = "First 6 rows of the participant info dataset")
First 6 rows of the participant info dataset
ID Exercise_Group Sex Age
P001 NONE Male 35
P002 Nonee Malee 57
P003 None Female 26
P004 None Female 29
P005 None Male 33
P006 None Female 33 
kable(head(sleep_data_midterm[, 1:4]), caption = "First 6 rows of the sleep dataset")
First 6 rows of the sleep dataset
ID Pre_Sleep Post_Sleep Sleep_Efficiency
P001 zzz-5.8 4.7 81.6
P002 Sleep-6.6 7.4 75.7
P003 NA 6.2 82.9
P004 SLEEP-7.2 7.3 83.6
P005 score-7.4 7.4 83.5
P006 Sleep-6.6 7.1 88.5

In this first chunk, I loaded all of my packages and uploaded the two datasets I will be working with.

Merge & Base Cleaning

unique(participant_info_midterm$Exercise_Group)
##  [1] "NONE"           "Nonee"          "None"           "N"             
##  [5] "Cardio"         "C"              "WEIGHTZ"        "WEIGHTS"       
##  [9] "WEIGHTSSS"      "Cardio+Weights" "CW"             "C+W"
participant_info_midterm <- participant_info_midterm %>%
  mutate(Exercise_Group = recode(Exercise_Group,
                                 "NONE" = "None",
                                 "Nonee" = "None",
                                 "N" = "None",
                                 "C" = "Cardio",
                                 "WEIGHTZ" = "Weights",
                                 "WEIGHTS" = "Weights",
                                 "WEIGHTSSS" = "Weights",
                                 "CW" = "Cardio+Weights",
                                 "C+W" = "Cardio+Weights"))

unique(participant_info_midterm$Sex)
## [1] "Male"    "Malee"   "Female"  "Femalee" "F"       "M"       "Fem"    
## [8] "MALE"    "Mal"
participant_info_midterm <- participant_info_midterm %>%
  mutate(Sex = recode(Sex,
                      "Malee" = "Male",
                      "Femalee" = "Female",
                      "F" = "Female",
                      "M" = "Male",
                      "Fem" = "Female",
                      "MALE" = "Male",
                      "Mal" = "Male"))

unique(sleep_data_midterm$Pre_Sleep)
##  [1] "zzz-5.8"   "Sleep-6.6" NA          "SLEEP-7.2" "score-7.4" "Sleep-6"  
##  [7] "zzz-8.1"   "sleep-5.5" "sleep-5.7" "score-7"   "Sleep-8"   "score-5.3"
## [13] "SLEEP-7.8" "zzz-6.7"   "SLEEP-7.4" "score-7.1" "zzz-6.8"   "Sleep-5.7"
## [19] "sleep-5"   "score-6"   "Sleep-6.2" "score-5.2" "sleep-6.6" "SLEEP-5.6"
## [25] "SLEEP-5.7" "Sleep-6.9" "SLEEP-6.9" "zzz-5.9"   "SLEEP-6.7" "zzz-5.4"  
## [31] "zzz-7"     "zzz-6"     "zzz-6.9"   "score-4"   "SLEEP-7.1" "SLEEP-5.8"
## [37] "sleep-6.7" "zzz-6.4"   "Sleep-6.5" "Sleep-7.2" "score-7.3" "sleep-6.2"
## [43] "sleep-7.3" "SLEEP-6.2" "SLEEP-6.6" "SLEEP-6.1" "score-6.7" "score-5.1"
## [49] "score-5.5" "sleep-6.4" "score-7.5" "zzz-6.6"   "Sleep-5.6" "SLEEP-5.1"
## [55] "sleep-6.5" "score-5.9" "score-6.6" "score-6.5" "sleep-8"   "score-6.2"
## [61] "zzz-7.3"   "sleep-5.6" "zzz-6.1"   "sleep-5.9" "SLEEP-4.4" "zzz-6.5"
sleep_data_midterm <- sleep_data_midterm %>%
  mutate(Pre_Sleep = str_remove(Pre_Sleep, "zzz-|SLEEP-|sleep-|score-|Sleep-"))

participant_sleep_midterm <- merge(sleep_data_midterm, participant_info_midterm, by="ID")

kable(head(participant_sleep_midterm[, 1:7]), caption = "First 6 rows of the merged dataset")
First 6 rows of the merged dataset
ID Pre_Sleep Post_Sleep Sleep_Efficiency Exercise_Group Sex Age
P001 5.8 4.7 81.6 None Male 35
P002 6.6 7.4 75.7 None Male 57
P003 NA 6.2 82.9 None Female 26
P004 7.2 7.3 83.6 None Female 29
P005 7.4 7.4 83.5 None Male 33
P006 6.6 7.1 88.5 None Female 33

In this chunk, I focused on cleaning the data. This included standardizing the ‘Exercise Group’ and ‘Sex’ columns as well as cleaning up the scores in the ‘Pre-Sleep’ column. Afterwards, I merged the two datasets into one large dataset.

Create Derived Variables

str(participant_sleep_midterm$Pre_Sleep)
##  chr [1:100] "5.8" "6.6" NA "7.2" "7.4" "6.6" "6" "8.1" "5.5" "5.7" "7" ...
participant_sleep_midterm <- participant_sleep_midterm %>%
  mutate(Pre_Sleep = as.numeric(Pre_Sleep))

participant_sleep_midterm$Sleep_Difference <- participant_sleep_midterm$Post_Sleep - participant_sleep_midterm$Pre_Sleep

participant_sleep_midterm <- participant_sleep_midterm %>%
  mutate(
    AgeGroup2 = case_when(
      Age < 40  ~ "<40",
      Age >= 40 ~ ">=40"
    )
  )

count(is.na(participant_sleep_midterm$Sleep_Difference))
## n_TRUE 
##     14
participant_sleep_midterm <- participant_sleep_midterm %>% 
  filter(!is.na(Sleep_Difference))

kable(head(participant_sleep_midterm[, 1:9]), caption = "Participant Sleep Data with Derived Variables")
Participant Sleep Data with Derived Variables
ID Pre_Sleep Post_Sleep Sleep_Efficiency Exercise_Group Sex Age Sleep_Difference AgeGroup2
P001 5.8 4.7 81.6 None Male 35 -1.1 <40
P002 6.6 7.4 75.7 None Male 57 0.8 >=40
P004 7.2 7.3 83.6 None Female 29 0.1 <40
P005 7.4 7.4 83.5 None Male 33 0.0 <40
P006 6.6 7.1 88.5 None Female 33 0.5 <40
P007 6.0 6.7 83.6 None Male 32 0.7 <40

In this chunk, I created new variables using existing data. I created a ‘Sleep_Difference’ variable by finding the difference between the Post- and Pre- sleep scores. I then created an age group variable by sorting participants into 2 groups: those under 40 years old, and those who are 40 years or older. Finally, I found that there were 14 rows that had NA values in the ‘Sleep_Difference’ column, and I deleted those participants from the dataset.

Descriptive Statistics

favstats(participant_sleep_midterm$Sleep_Difference)
##   min  Q1 median  Q3 max      mean        sd  n missing
##  -1.1 0.3   0.75 1.1 2.1 0.6825581 0.6610494 86       0
favstats(participant_sleep_midterm$Sleep_Efficiency)
##   min     Q1 median     Q3   max     mean       sd  n missing
##  71.7 79.975   83.3 88.425 101.5 83.77558 5.973804 86       0
participant_sleep_midterm %>%
  group_by(Exercise_Group) %>%
  summarize(
    mean_Sleep_Difference = mean(Sleep_Difference),
    mean_Sleep_Efficiency = mean(Sleep_Efficiency)
  )
## # A tibble: 4 × 3
##   Exercise_Group mean_Sleep_Difference mean_Sleep_Efficiency
##   <chr>                          <dbl>                 <dbl>
## 1 Cardio                        1.14                    85.4
## 2 Cardio+Weights                0.861                   86.8
## 3 None                          0.0476                  81.1
## 4 Weights                       0.667                   81.5

This is a summary of the descriptive statistics of the ‘Sleep_Difference’ and ’Sleep_Efficiency variables, including mean, sd, min, max, and group-wise means.

Visualizations

ggplot(participant_sleep_midterm, aes(x = Sleep_Difference, y = Exercise_Group, fill = Exercise_Group)) +
  geom_boxplot(outlier.shape = NA) +
  theme_classic() +
  coord_flip() +
  theme(legend.position = "none") +
  labs(title = "Difference in Sleep Efficiency by Type of Exercise",
          x = "Difference in Sleep Efficiency", y = "Type of Exercise")

ggplot(participant_sleep_midterm, aes(x = Sleep_Efficiency, y = Exercise_Group, fill = Exercise_Group)) +
  geom_boxplot(outlier.shape = NA) +
  theme_classic() +
  coord_flip() +
  theme(legend.position = "none") +
  labs(title = "Sleep Efficiency by Type of Exercise",
       x = "Sleep Efficiency", y = "Type of Exercise")

ggplot(participant_sleep_midterm, aes(x = Sleep_Difference, y = Sleep_Efficiency)) +
  geom_point() +
  theme_classic() +
  geom_smooth(method = "lm", color = "steelblue") +
  labs(title = "Difference in Sleep Efficiency by Participant After Exercise",
       x = "Difference in Sleep Efficiency", y = "Sleep Efficiency")
## `geom_smooth()` using formula = 'y ~ x'

Here are three visualizations: two box plots that represent how type of exercise affects difference in sleep efficiency and sleep efficiency itself, and a scatterplot that represents how exercise affected the difference in sleep efficiency for each participant (including a trend line).

T-Tests

Independent T-Test: Sleep_Difference ~ Sex

sleepdif_sex_t_test <-t.test(Sleep_Difference ~ Sex, data = participant_sleep_midterm)
sleepdif_sex_t_test
## 
##  Welch Two Sample t-test
## 
## data:  Sleep_Difference by Sex
## t = 1.5801, df = 77.647, p-value = 0.1182
## alternative hypothesis: true difference in means between group Female and group Male is not equal to 0
## 95 percent confidence interval:
##  -0.05865017  0.50972574
## sample estimates:
## mean in group Female   mean in group Male 
##            0.7795918            0.5540541

Female mean: 0.7796

Male mean: 0.5541

P-value: 0.12 (>0.05)

Differences in sleep efficiency between male and female participants are not statistically different from one another/not statistically significant. We fail to reject the null.

Independent T-Test: Sleep_Difference ~ AgeGroup2

sleepdif_age_t_test <-t.test(Sleep_Difference ~ AgeGroup2, data = participant_sleep_midterm)
sleepdif_age_t_test
## 
##  Welch Two Sample t-test
## 
## data:  Sleep_Difference by AgeGroup2
## t = -1.3746, df = 36.662, p-value = 0.1776
## alternative hypothesis: true difference in means between group <40 and group >=40 is not equal to 0
## 95 percent confidence interval:
##  -0.50676303  0.09717936
## sample estimates:
##  mean in group <40 mean in group >=40 
##          0.6373134          0.8421053

<40 mean: 0.6373

=>40 mean: 0.8421

P-value: 0.18 (>0.05)

Differences in sleep efficiency between participants under 40 years old and participants 40 years old or older are not statistically different from one another/not statistically significant. Again, we fail to reject the null.

ANOVAs

ANOVA A: Sleep_Difference ~ Exercise_Group

anova_model_diff_exercise <- aov(Sleep_Difference ~ Exercise_Group, data = participant_sleep_midterm)
summary(anova_model_diff_exercise)
##                Df Sum Sq Mean Sq F value   Pr(>F)    
## Exercise_Group  3  13.56   4.520   15.72 3.67e-08 ***
## Residuals      82  23.58   0.288                     
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
supernova(anova_model_diff_exercise)
##  Analysis of Variance Table (Type III SS)
##  Model: Sleep_Difference ~ Exercise_Group
## 
##                              SS df    MS      F   PRE     p
##  ----- --------------- | ------ -- ----- ------ ----- -----
##  Model (error reduced) | 13.560  3 4.520 15.717 .3651 .0000
##  Error (from model)    | 23.583 82 0.288                   
##  ----- --------------- | ------ -- ----- ------ ----- -----
##  Total (empty model)   | 37.144 85 0.437

F-value: 15.717 (large)

df: 85

p-value: 3.67e-08 (<0.05, significant!)

PRE: 0.3651 (~36.6% of variance)

Exercise group accounts for about 37% of total variance in difference in sleep efficiency.

Post-Hoc: Tukey HSD

TukeyHSD(anova_model_diff_exercise)
##   Tukey multiple comparisons of means
##     95% family-wise confidence level
## 
## Fit: aov(formula = Sleep_Difference ~ Exercise_Group, data = participant_sleep_midterm)
## 
## $Exercise_Group
##                              diff        lwr         upr     p adj
## Cardio+Weights-Cardio  -0.2772257 -0.7017134  0.14726203 0.3237562
## None-Cardio            -1.0904762 -1.5245041 -0.65644825 0.0000000
## Weights-Cardio         -0.4714286 -0.9054565 -0.03740063 0.0278779
## None-Cardio+Weights    -0.8132505 -1.2377382 -0.38876282 0.0000171
## Weights-Cardio+Weights -0.1942029 -0.6186906  0.23028480 0.6287294
## Weights-None            0.6190476  0.1850197  1.05307556 0.0018927

The ‘Cardio’ and ‘Cardio+Weights’ groups were tied for the most positive change in sleep quality; there was no statistically significant difference between these two exercise groups. In addition, the ‘Weights’ group showed a significantly greater improvement in sleep quality compared to the ‘None’ group. The ‘None’ group produced significantly worse results in sleep quality than all of the other groups.

ANOVA B: Sleep_Efficiency ~ Exercise_Group

anova_model_eff_exercise <- aov(Sleep_Efficiency ~ Exercise_Group, data = participant_sleep_midterm)
summary(anova_model_eff_exercise)
##                Df Sum Sq Mean Sq F value  Pr(>F)   
## Exercise_Group  3  540.4   180.1   5.925 0.00104 **
## Residuals      82 2492.9    30.4                   
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
supernova(anova_model_eff_exercise)
##  Analysis of Variance Table (Type III SS)
##  Model: Sleep_Efficiency ~ Exercise_Group
## 
##                                SS df      MS     F   PRE     p
##  ----- --------------- | -------- -- ------- ----- ----- -----
##  Model (error reduced) |  540.400  3 180.133 5.925 .1782 .0010
##  Error (from model)    | 2492.939 82  30.402                  
##  ----- --------------- | -------- -- ------- ----- ----- -----
##  Total (empty model)   | 3033.339 85  35.686

F-value: 5.925

df: 85

p-value: 0.001 (<0.05, significant!)

PRE: 0.1782 (~17.8% of variance)

Exercise group accounts for about 18% of total variance in sleep efficiency.

Post-Hoc: Tukey HSD

TukeyHSD(anova_model_eff_exercise)
##   Tukey multiple comparisons of means
##     95% family-wise confidence level
## 
## Fit: aov(formula = Sleep_Efficiency ~ Exercise_Group, data = participant_sleep_midterm)
## 
## $Exercise_Group
##                              diff        lwr         upr     p adj
## Cardio+Weights-Cardio   1.3871636  -2.977172  5.75149915 0.8383629
## None-Cardio            -4.3761905  -8.838613  0.08623232 0.0566544
## Weights-Cardio         -3.9904762  -8.452899  0.47194661 0.0962888
## None-Cardio+Weights    -5.7633540 -10.127690 -1.39901844 0.0046379
## Weights-Cardio+Weights -5.3776398  -9.741975 -1.01330416 0.0094267
## Weights-None            0.3857143  -4.076709  4.84813708 0.9958617

Only the ‘Cardio+Weights’ group showed statistically significant improvements in sleep efficiency compared to other groups. Specifically, it performed significantly better than both the ‘Weights’ and ‘None’ groups. There was no significant difference between the ‘Cardio’ and ‘Cardio+Weights’ groups. The ‘Cardio’ group showed potential improvement when compared to the ‘None’ group, but the difference was not found to be statistically significant. The ‘None’ group had the worst overall sleep efficiency.

The ‘Cardio+Weights’ group

Synthesis & Recommendation

Based on both Sleep_Difference and Sleep_Efficiency, the ‘Cardio+Weights’ exercise is the best choice for improving sleep. It had the highest average improvements and was the only group that showed statistically significant improvements with both variables. For Sleep_Difference, the ANOVA showed a strong effect (F(3, 82) = 15.72, p < .001), with ‘Cardio+Weights’ showing significant improvement compared to the ‘None’ group (p < .001) and ‘Weights’ group (p = .00002). For Sleep_Efficiency, the ANOVA model was also significant (F(3, 82) = 5.93, p = .001), and again, ‘Cardio+Weights’ had much better quality than ‘None’ (p = .0046) and ‘Weights’ (p = .0094). While ‘Cardio’ alone came close to significance, it was not significantly better than any other group. Overall, ‘Cardio+Weights’ consistently led to better sleep outcomes and is the most effective recommendation.

Reflection

The analyses were definitely the toughest part- coding them was one thing, but understanding and interpreting them required a completely different mindset (I need to re-up on my statistics for sure). I would say that took the longest. I felt really confident about the cleaning, merging, and plotting, and I had fun trying out different ggplot2 themes to see which I liked the best. Though by the end, I felt confident enough in my data analyses and interpretations, it took a lot of resources to bring me to those conclusions, and I would like to work on my interpretation skills for the future so that it does not take so long for me to parse through everything to figure out what it all means.