-
Notifications
You must be signed in to change notification settings - Fork 48
/
Module-6-Overall-Example.R
205 lines (102 loc) · 4.66 KB
/
Module-6-Overall-Example.R
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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
#########################################
##
## Module 6 Live Office Hour
## Katherine Grzesik, 3/3/19
##
#########################################
# install.packages("aod")
# install.packages("pROC")
library(aod)
library(pROC)
## Creating a fake data set of men and women with how much they spent on gas
## and whether or not they purchases a slice of pizza
sex=rep(c(1,2),each=50)
set.seed(38281)
data<-data.frame(sex=sex,
gas=c(rnorm(50,12,2),rnorm(50,20,3)),
pizza=factor(c(rbinom(50,1,.50),rbinom(50,1,.20)),levels = c(0,1),labels=c("No","Yes")))
str(data)
data$sexDummy<-ifelse(data$sex==2,1,0)
data$sexFactor<-factor(ifelse(data$sex==2,"F","M"))
str(data)
## Let's take a look at the data
summary(data)
tt<-table(Pizza=data$pizza, Sex=data$sexDummy)
prop.male.pizza<-tt[2,1]/colSums(tt)[1]
prop.female.pizza<-tt[2,2]/colSums(tt)[2]
abs(prop.male.pizza-prop.female.pizza)
## Do the same proportion of men and women buy pizza when getting gas?
prop.test(x=tt["Yes",], n=colSums(tt), correct=FALSE)
## Do women buy pizza more or less frequently?
# Ha: first group prop > second group prop
prop.test(x=tt["Yes",], n=colSums(tt), correct=FALSE, alternative = "greater")
# Ha: first group prop < second group prop
prop.test(x=tt["Yes",], n=colSums(tt), correct=FALSE, alternative = "less")
## Chi-Square
sum((Exp - Obs)^2/Exp))
where Expectation assumes independence
addmargins(tt)
## Can we predict whether a person will get pizza based on their sex?
str(data)
m<-glm(pizza ~ sexDummy, data=data, family = binomial)
summary(m)
coef(m)
exp(coef(m)) # Females are less likely to get pizza since < 1
1/.1387 # Males are 7.2 times as likely as females to get pizza!
exp(confint.default(m))
## Reference Levels!
# If numeric, lowest number is reference
# If factor, defaults to alphabetical UNLESS you tell it which to use.
data$sexFactorFlipped<-relevel(data$sexFactor,ref = "M") # Makes MALE as reference
m.factor<-glm(pizza ~ sexFactor, data=data, family = binomial)
m.flipped<-glm(pizza ~ sexFactorFlipped, data=data, family = binomial)
coef(m.factor)
coef(m.flipped)
exp(confint.default(m.factor))
## Check how well the model predicts pizza purchasing
data$ProbofPizza<-predict(m, type="response") # Vector of P(getting pizza)
(g<-roc(data$pizza~data$ProbofPizza))
# C-statistics = 0.7228
## Create the ROC curve
plot(g, main="AUC=0.7228")
## Does how much a person spends on gas impact the pizza purchase?
## How does the sex of a person predict pizza purchase after adjusting for
# how much they spent in gas?
str(data)
m1<-glm(pizza~sexDummy + gas, data=data, family=binomial)
summary(m1)
exp(coef(m1))
1/.0628 # Males are ~16 times as likely to buy pizza than females, after accounting
# for the amount spent on gas
# Can only do this flip trick on DICHOTOMOUS CATEGORICAL VARIABLES
## What about for every 10 cents spent on gas? *Going from 1 penny to 1 dime
coef(m1)
exp(coef(m1)["gas"]*10)
exp((m1$coefficients[3]-qnorm(.975)*summary(m1)$coefficients[3,2])*10)
exp((m1$coefficients[3]+qnorm(.975)*summary(m1)$coefficients[3,2])*10)
confint.default(m1)
exp(confint.default((m1))*10) # Yes, appropriate way to calculate them FOR ONLY the gas variable
## What about the ROC Curve and the C-Statistic now?
data$ProbofPizza2<-predict(m1, type="response")
(g2<-roc(data$pizza ~ data$ProbofPizza2))
plot(g2, main="AUC=0.7393")
plot(1-g2$specificities, g2$sensitivities, type="line")
g2$thresholds
data$Nonsense<-sample(c(0,1),nrow(data), replace=TRUE)
data$Nonsense2<-sample(c(0,1),nrow(data), replace=TRUE)
m3<-glm(pizza~Nonsense + Nonsense2, data=data, family=binomial)
summary(m3)
wald.test(b=coef(m3), Sigma=vcov(m3), Terms=2:3)
wald.test(b=coef(m), Sigma=vcov(m), Terms=2) # .000069
wald.test(b=coef(m1), Sigma=vcov(m1), Terms=2:3) # .00032 <- model got worse, marginally so
## Extra Content - if time
## Cutoffs, Sensitivity, Specificity
## Source for these terms: https://www.med.emory.edu/EMAC/curriculum/diagnosis/sensand.htm
# Sensitivity: Probabilty of Detecting True Positive
# Sensitivity= true positives/(true positive + false negative)
# Specificity: Probability of Detecting a True Negative
# Specificity=true negatives/(true negative + false positives)
# Temp is a continuous variable, need to use temp_level
# Logisitic Regression REQUIRES a categorical outcome
> Q3.m3 <- glm(data.heart$temp ~ data.heart$girls, data = data.heart, family=binomial)
Error in eval(family$initialize) : y values must be 0 <= y <= 1