-
Notifications
You must be signed in to change notification settings - Fork 0
/
pres-script-yscale-means.Rmd
181 lines (139 loc) · 7.82 KB
/
pres-script-yscale-means.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
---
title: "pres-script"
author: "Sebastian Wagner"
date: "2023-06-08"
output: pdf_document
---
```{r hidden=TRUE}
# for pretty printing
library(glue)
```
## Data Preparation
```{r}
# set working dir to where data is
#setwd("INSERT")
# Imported preprocessed data
dat <- read.csv("Data_clean/log_p6.txt", sep="|")
# Filter sitting data out
sitting_phone <- dat[dat$statusId == 2, ] # this is the id for sitting on phone
sitting_no_phone <- dat[dat$statusId == 1, ] # this is the id for sitting
walking_phone <- dat[dat$statusId == 4, ] # this is the id for walking on phone
walking_no_phone <- dat[dat$statusId == 3, ] # this is the id for walking
## Timestamp = milliseconds after experiment start
mt_1 <- min(sitting_phone$timestamp)
sitting_phone$timestamp <- sitting_phone$timestamp - mt_1
mt_2 <- min(sitting_no_phone$timestamp)
sitting_no_phone$timestamp <- sitting_no_phone$timestamp - mt_2
mt_3 <- min(walking_phone$timestamp)
walking_phone$timestamp <- walking_phone$timestamp - mt_3
mt_4 <- min(walking_no_phone$timestamp)
walking_no_phone$timestamp <- walking_no_phone$timestamp - mt_4
# Acceleration level can be calculated with this:
sitting_phone$acceleration <- sqrt(sitting_phone$x^2 + sitting_phone$y^2 + sitting_phone$z^2)
sitting_no_phone$acceleration <- sqrt(sitting_no_phone$x^2 + sitting_no_phone$y^2 + sitting_no_phone$z^2)
walking_phone$acceleration <- sqrt(walking_phone$x^2 + walking_phone$y^2 + walking_phone$z^2)
walking_no_phone$acceleration <- sqrt(walking_no_phone$x^2 + walking_no_phone$y^2 + walking_no_phone$z^2)
sitting_phone <- sitting_phone[c("timestamp", "acceleration")]
sitting_no_phone <- sitting_no_phone[c("timestamp", "acceleration")]
walking_phone <- walking_phone[c("timestamp", "acceleration")]
walking_no_phone <- walking_no_phone[c("timestamp", "acceleration")]
# Calculate the common y-axis limits
y_limits <- c(min(sitting_phone$acceleration, walking_phone$acceleration, sitting_no_phone$acceleration, walking_no_phone$acceleration),
max(sitting_phone$acceleration, walking_phone$acceleration, sitting_no_phone$acceleration, walking_no_phone$acceleration))
# Calculate the common x-axis limits
x_limits <- c(min(sitting_phone$timestamp, walking_phone$timestamp, sitting_no_phone$timestamp, walking_no_phone$timestamp),
max(sitting_phone$timestamp, walking_phone$timestamp, sitting_no_phone$timestamp, walking_no_phone$timestamp))
# Plot on phone data
plot(sitting_phone$timestamp,sitting_phone$acceleration,type = "l",col = 2,xlab = "Timestamp",ylab = "Acceleration sitting on phone", ylim = y_limits, xlim = x_limits)
plot(walking_phone$timestamp,walking_phone$acceleration,type = "l",col = 4,xlab = "Timestamp",ylab = "Acceleration walking on phone", ylim = y_limits, xlim = x_limits)
# Plot off phone data
plot(sitting_no_phone$timestamp,sitting_no_phone$acceleration,type = "l",col = 2,xlab = "Timestamp",ylab = "Acceleration walking not on phone", ylim = y_limits, xlim = x_limits)
plot(walking_no_phone$timestamp,walking_no_phone$acceleration,type = "l",col = 4,xlab = "Timestamp",ylab = "Acceleration sitting not on phone", ylim = y_limits)
# We will minimize the effects of starting phase on the statistical analysis by
# eliminating the first 3000 timestamps:
sitting_phone <- sitting_phone[-(1:3000),]
sitting_no_phone <- sitting_no_phone[-(1:3000),]
walking_phone <- walking_phone[-(1:3000),]
walking_no_phone <- walking_no_phone[-(1:3000),]
# Define means
mean_sitting_phone <- mean(sitting_phone$acceleration)
mean_sitting_no_phone <- mean(sitting_no_phone$acceleration)
mean_walking_phone <- mean(walking_phone$acceleration)
mean_walking_no_phone <- mean(walking_no_phone$acceleration)
# Create a data frame from the means
means_dataset <- data.frame(
Condition = c("Sitting with Phone", "Sitting without Phone", "Walking with Phone", "Walking without Phone"),
Mean_Acceleration = c(mean_sitting_phone, mean_sitting_no_phone, mean_walking_phone, mean_walking_no_phone)
)
# Print the dataset
means_dataset
```
## Testing
> We expect to observe lower acceleration levels during sitting while using the phone compared to sitting without phone usage.
> We expect to observe higher acceleration levels during walking while using the phone compared to walking without phone usage.
```{r}
# Plot on phone data
plot(sitting_phone$timestamp,sitting_phone$acceleration,type = "l",col = 2,xlab = "Timestamp",ylab = "Acceleration")
# Plot off phone data
plot(sitting_no_phone$timestamp,sitting_no_phone$acceleration,type = "l",col = 2,xlab = "Timestamp",ylab = "Acceleration no phone")
# Plot acceleration frequency
hist(sitting_phone$acceleration, breaks=500)
hist(sitting_no_phone$acceleration, breaks=500)
hist(walking_phone$acceleration, breaks=500)
hist(walking_no_phone$acceleration, breaks=500)
# We can see that our data does not necessarily follow a normal distribution
# This means we use Mann-Whitney test
wilcox.test(sitting_phone$acceleration, sitting_no_phone$acceleration)
# Sitting while not on the phone has less acceleration
mean(sitting_phone$acceleration)
mean(sitting_no_phone$acceleration)
wilcox.test(walking_phone$acceleration, walking_no_phone$acceleration)
# Walking while not on the phone has less acceleration
median(walking_phone$acceleration)
median(walking_no_phone$acceleration)
# Let's see what the p value of testing the same class would be
wilcox.test(walking_phone$acceleration[1:15000], walking_phone$acceleration[15000:30000])
```
## Multiple Testing
```{r}
# We've cramped the procedure in this function:
test_participant <- function(filename) {
dat <- read.csv(filename, sep="|")
sitting_phone <- dat[dat$statusId == 2, ] # this is the id for sitting on phone
sitting_no_phone <- dat[dat$statusId == 1, ] # this is the id for sitting
walking_phone <- dat[dat$statusId == 4, ] # this is the id for walking on phone
walking_no_phone <- dat[dat$statusId == 3, ] # this is the id for walking
sitting_phone <- sitting_phone[-(1:3000),]
sitting_no_phone <- sitting_no_phone[-(1:3000),]
walking_phone <- walking_phone[-(1:3000),]
walking_no_phone <- walking_no_phone[-(1:3000),]
sitting_phone$acceleration <- sqrt(sitting_phone$x^2 + sitting_phone$y^2 + sitting_phone$z^2)
sitting_no_phone$acceleration <- sqrt(sitting_no_phone$x^2 + sitting_no_phone$y^2 + sitting_no_phone$z^2)
walking_phone$acceleration <- sqrt(walking_phone$x^2 + walking_phone$y^2 + walking_phone$z^2)
walking_no_phone$acceleration <- sqrt(walking_no_phone$x^2 + walking_no_phone$y^2 + walking_no_phone$z^2)
# Print p-value of test
sitting_res = glue('Sitting p = {wilcox.test(sitting_phone$acceleration, sitting_no_phone$acceleration)$p.value}')
walking_res = glue('Walking p = {wilcox.test(walking_phone$acceleration, walking_no_phone$acceleration)$p.value}')
glue('{sitting_res}\n{walking_res}')
}
# Test all participants
test_participant("Data_clean/log_p1.txt")
test_participant("Data_clean/log_p2.txt")
test_participant("Data_clean/log_p3.txt")
test_participant("Data_clean/log_p4.txt")
test_participant("Data_clean/log_p5.txt")
test_participant("Data_clean/log_p6.txt")
test_participant("Data_clean/log_p7.txt")
test_participant("Data_clean/log_p8.txt")
test_participant("Data_clean/log_p9.txt")
test_participant("Data_clean/log_p10.txt")
# All of the p values are very low and fall below 5 percent threshold
# Even when you corrent it with Bonferroni correction the threshold for each
# test is alpha / 10 = 0.005 which they are still below
```
Now, for the feasibility study we have 10 recordings from different people that we'll use.
In the following we will perform a test for all of the data combined.
We'll be doing the above procedure for each participants data.
In the end this will amount to 10 * 3 = 30 F-tests for each participant and dimension.
For the test we will arbitrarily set our alpha to 5% and correct it with the Bonferroni correction.
We divide it by the number of tests performed, to get an acceptable alpha of 0.05 / 30 = 0.00167 per test.