R Notebook

setwd("~/Documents/Winter 2018/Statistics 2")
library(tidyverse)
library(tidyr)
library(haven)
info <- read_dta("~/Documents/Winter 2018/Statistics 2/afghan_fraud.DTA")
install.packages("margins")

Error in install.packages : Updating loaded packages

library("margins")

fraud_lpm = lm(fraud ~ election_viol+ num_closed_polls+ dist2kabul+ elevation,data = info)
vals <- fraud_lpm$fitted.values
range(vals)

[1] 0.05417908 1.40588290

summary(fraud_lpm)

Call:
lm(formula = fraud ~ election_viol + num_closed_polls + dist2kabul + 
    elevation, data = info)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.80772 -0.17030 -0.12534 -0.07245  0.94156 

Coefficients:
                  Estimate Std. Error t value Pr(>|t|)    
(Intercept)      4.061e-02  6.688e-02   0.607   0.5441    
election_viol    4.583e-01  2.364e-01   1.938   0.0533 .  
num_closed_polls 5.717e-03  9.669e-04   5.912 7.49e-09 ***
dist2kabul       1.151e-05  1.159e-04   0.099   0.9209    
elevation        3.386e-05  2.258e-05   1.499   0.1346    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.3672 on 382 degrees of freedom
  (4 observations deleted due to missingness)
Multiple R-squared:  0.1115,    Adjusted R-squared:  0.1022 
F-statistic: 11.99 on 4 and 382 DF,  p-value: 3.472e-09

# 2a: An increase in the unit of the measure of violence in the district in 2009 is associated with a 4.583e01 increase in the probability of election fraud. 
# 2b: See model created. The range of this variability is from 0.05 to 1.41. This is problematic because probability cannot be greater than 1.  

fraud_probit <- glm(fraud ~ election_viol+ num_closed_polls+ dist2kabul+ elevation,data = info)
summary(fraud_probit)

Call:
glm(formula = fraud ~ election_viol + num_closed_polls + dist2kabul + 
    elevation, data = info)

Deviance Residuals: 
     Min        1Q    Median        3Q       Max  
-0.80772  -0.17030  -0.12534  -0.07245   0.94156  

Coefficients:
                  Estimate Std. Error t value Pr(>|t|)    
(Intercept)      4.061e-02  6.688e-02   0.607   0.5441    
election_viol    4.583e-01  2.364e-01   1.938   0.0533 .  
num_closed_polls 5.717e-03  9.669e-04   5.912 7.49e-09 ***
dist2kabul       1.151e-05  1.159e-04   0.099   0.9209    
elevation        3.386e-05  2.258e-05   1.499   0.1346    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for gaussian family taken to be 0.134841)

    Null deviance: 57.974  on 386  degrees of freedom
Residual deviance: 51.509  on 382  degrees of freedom
  (4 observations deleted due to missingness)
AIC: 329.81

Number of Fisher Scoring iterations: 2

# 2c: The range is from 0 to 1.. 
# the sign of the coefficient of the number of polling stations in lpm is positive. 
# it is still positive in the probit model. The two values are the same. This is expected 
# because the linear model is also measuring the effect of a variable on a binary variable. 
# 2d: 

# e
avg_violence <- mean(info$election_viol, na.rm=TRUE)
partial_effect <- 4.583e-01 * avg_violence
View(avg_violence)
 # therefor the partial affect is very small 

# f
marg <- margins(fraud_probit)

Error in margins(fraud_probit) : could not find function "margins"