Bentley - Project

Sources

There are a number of sources that deal with these questions. I found the following three to be the most interesting:

• https://www.govtrack.us/congress/bills/statistics

• http://www.uscourts.gov/sites/default/files/allauth.pdf

• https://voteview.com/data

VoteView has the most interesting data to work with, particularly:

• President Support Score which reflects the fraction of times that each Congress voted to support the president’s position across all votes on which the president’s position can be inferred.

pss <- read.csv("https://raw.githubusercontent.com/ajbentley/cuny_ms_ds/master/606/presidential_support_summary.csv", header = TRUE, stringsAsFactors = FALSE)

knitr::kable(head(pss), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

• NOMINATE which measures the political leanings of individuals or bodies on two liberal-conservative dimensions: economic and social.

I’ll be bringing in two files that feature NOMINATE info. The first records every Roll Call Vote (RCV) Congress has made.

rcv <- read.csv("https://raw.githubusercontent.com/ajbentley/cuny_ms_ds/master/606/rollcalls.csv", header = TRUE, stringsAsFactors = FALSE)

knitr::kable(head(rcv), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

The second NOMINATE file records the NOMINATE standing of every president and member of congress.

members <- read.csv("https://raw.githubusercontent.com/ajbentley/cuny_ms_ds/master/606/HSall_members.csv", header = TRUE, stringsAsFactors = FALSE)

knitr::kable(head(members), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

Transformations

rcv$date <- as.character(mdy(rcv$date))
rcv$year <- str_extract(rcv$date, "[:digit:]{4}")

# creating fails and success by year
# creating new df with just year and success counts

# rcv_annual_successes <- select(rcv, c("year", "Success")) %>% 
#  group_by(year) %>% 
#  count(rcv_annual_successes$Success) 

# rcv_ann_succ <- rcv_annual_successes

# rcv_ann_succ$Success <- as.factor(rcv_ann_succ$Success)

# rcv_ann_succ$Success <- revalue(rcv_ann_succ$Success, c("0" = "Fail", "1" = "Succeed"))

# rcv_ann_succ1 <- spread(rcv_ann_succ, Success, freq)

# rcv_ann_succ1$succ.to.fail.ratio <- rcv_ann_succ1$Succeed / rcv_ann_succ1$Fail

# write.csv(rcv_ann_succ1, "rcv_ann_succ1.csv")

rcv_ann_succ1 <- read.csv("https://raw.githubusercontent.com/ajbentley/cuny_ms_ds/master/606/rcv_ann_succ1.csv", header = TRUE, stringsAsFactors = FALSE) %>% 
  select(-"X")

rcv_ann_succ1$year <- as.character(rcv_ann_succ1$year)

rcv_annual_nominate <- select(rcv, c("year", "nominate_mid_1", "nominate_mid_2")) %>% 
  group_by(year)%>% 
  summarise_all(funs(mean))

colnames(rcv_annual_nominate) <- c("year", "congress_mid_dim_1", "congress_mid_dim_2")

rcv2 <- left_join(rcv_ann_succ1, rcv_annual_nominate, by = "year")

# create new df that is just the President and his nom score

presidents <- filter(members, chamber == "President") %>%
  select("bioname", "nominate_geo_mean_probability")

# rename columns
colnames(presidents) <- c("President", "NGMP")

# change nom score to character type
presidents$NGMP <- as.character(presidents$NGMP)

# exclude Washington / Adams.
pns <-slice(presidents, 7:123)

# fill down the nom score where there is a repeated line
pns$NGMP <- na.locf(pns$NGMP)

pl <- read.csv("https://gist.githubusercontent.com/namuol/2657233/raw/74135b2637e624848c163759be9cd14ae33f5153/presidents.csv", header = TRUE, stringsAsFactors = FALSE)

knitr::kable(head(pl), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

# selecting desired columns only
pl2 <- select(pl, "President", "Party", "Took.office", "Left.office")

# converting N/A for Obama's last day to 2017/01/19
pl2[44,3] <- "20/01/2017"

# adding trump
trump <- c("Donald Trump", "Republican", "20/01/2017", "20/01/2020")

pl2 <- rbind(pl2, trump)

# create new cols for term start year and term end year
pl2$term.start.yr <- as.numeric(str_extract(pl2$Took.office, "[:digit:]{4}"))
pl2$term.end.yr <- as.numeric(str_extract(pl2$Left.office, "[:digit:]{4}"))

# subtacting one year from Left office so that there is no overlap between administrations.
pl2$term.end.yr <- pl2$term.end.yr - 1


knitr::kable(head(pl2), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

pPn <- c("WASHINGTON, George", "ADAMS, John", unique(pns$President[1:50]), unique(pns$President[51:117]))

pl2$President <- pPn

# create new df with presidents from Eisenhower on
pl3 <- pl2[34:45, 1:6]

# create new df with years we're looking for and assign it a column name of yr
year <- data.frame(seq(1953, 2019))

colnames(year) <- "year"

# create column in pl3 with the same information but new name of start year (to match yr df.
pl3$year <- pl3$term.start.yr

# join pl3 and year
pres_year <- left_join(year, pl3, by = "year")

knitr::kable(head(pres_year), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

# fill down presidents name and party
pres_year$President <- na.locf(pres_year$President)
pres_year$Party <- na.locf(pres_year$Party)

# keep just the two columns needed
pres_yr <- pres_year[1:67, 1:3]

knitr::kable(head(pres_yr), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

pss <- left_join(pres_yr, pss, by = "year")

df <- pss[3:62, 1:6]

knitr::kable(head(df), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

rcv2$year <- as.numeric(rcv2$year)

df <- left_join(df, rcv2, by = "year")

colnames(df) <- c("year", "President", "Party", "House.PSS", "Senate.PSS", "Avg.PSS", "Leg.Fail", "Leg.Succ", "Leg.Succ.to.Fail.Ratio", "Cong.Dim1", "Cong.Dim2")

knitr::kable(head(df), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

President support score

This file has four pieces of information:

• Year
• Presidential Support Score (PSS) of the House of Representatives
• PSS of the Senate
• Average of the two PSS scores

knitr::kable(head(pss), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

What are the summary statistics for each column?

summary(pss)

##       year       President            Party               house      
##  Min.   :1953   Length:67          Length:67          Min.   :36.50  
##  1st Qu.:1970   Class :character   Class :character   1st Qu.:47.75  
##  Median :1986   Mode  :character   Mode  :character   Median :53.55  
##  Mean   :1986                                         Mean   :52.81  
##  3rd Qu.:2002                                         3rd Qu.:58.88  
##  Max.   :2018                                         Max.   :64.50  
##                                                       NA's   :7      
##      senate           avg       
##  Min.   :46.50   Min.   :45.10  
##  1st Qu.:57.42   1st Qu.:53.02  
##  Median :61.30   Median :58.25  
##  Mean   :62.86   Mean   :57.86  
##  3rd Qu.:67.53   3rd Qu.:61.95  
##  Max.   :81.70   Max.   :70.20  
##  NA's   :7       NA's   :7

Generally speaking the Senate has been on the side of the President more often than the House has.

Let’s take a look at the distributions of each.

par(mfrow = c(2, 2))
set.seed(23)

hist(pss$house, breaks = 8)
hist(pss$senate, breaks = 8)

housepss_means <- rep(pss$house, 50)

for(i in pss$house){
   hpss_means <- sample(pss$house, 15, replace = FALSE)
   housepss_means[i] <- mean(hpss_means)
   }

hist(housepss_means, breaks = 8)

senatepss_means <- rep(pss$senate, 50)

for(i in pss$senate){
   spss_means <- sample(pss$senate, 15, replace = FALSE)
   senatepss_means[i] <- mean(spss_means)
   }

hist(senatepss_means, breaks = 8)

The Senate looks like it is fairly normal (with a slight right skew) both in its overall distribution and in the distribution of its means.

The House, on the other hand, looks somewhat left-skewed. This is contrary to what we saw in the means as it suggests that the Presidential Support Score is more often higher in the House.

It’s always frustrating when two analyses conflict with each other but we’ll look at some inferences later that might clear things up.

Has this changed over time?

This is an interactive chart: hover over a data point to see the Senate and House PSS as well as the year.
Administrations start in the year after the election so while so while Reagan was elected in 1980 his administration started in 1981.

a <- list(
  autotick = FALSE,
  ticks = "outside",
  tick0 = 1961,
  dtick = 4
  )

yt <- list(title = "Presidential Support Score")

sby_tl <- plot_ly(pss, x = ~year, y = ~house, name = "House PSS", type = 'scatter', mode = 'lines') %>%
  add_trace(y = ~senate, name = "Senate PSS", mode = 'lines+markers') %>% 
  layout(xaxis=a, yaxis = yt, hovermode = 'compare')

sby_tl

While the House has fairly consistently been less in line with the President the disparity has at times been far greater than ever before in recent years.

Let’s take a look at that difference over time.

pss$diff <- abs(pss$house - pss$senate)

pss_diff_tl <- plot_ly(pss, x = ~year, y = ~diff, name = "PSS Difference", type = 'scatter', mode = 'lines')

pss_diff_tl

Wow…I really wasn’t expecting the visual to be so stark in the rise in the difference between the two bodies.

• Prior to 1981 (Reagan’s first year) the average difference was 4.9 points.
  
• From 1981 - 1992 (Reagan and Bush 1) the average difference was 13.3 points.
  
• From 1993 - 2000 (Clinton) the average difference was 10.0 points.
  
• From 2001 - 2008 (Bush 2) the average difference was 17.0 points.
  
• From 2009 - 2012 (Obama’s first term) the average difference was 19.2 points.
  
• From 2013 - 2014 (the latest year available / Obama’s first term) the average difference was 28.9 points.
  

Since 1970 there have only been three periods when the two bodies were within 3 points of each other:

• 1970 - 1973 (most of Nixon’s term): average 0.5 difference.
  
• 1980 (Carter’s last year): 1.2 difference
  
• 1992 - 1993 (Bush 1’s last year and Clinton’s first year): average 1.2 difference (Bush 2.5 and Clinton 0.2, the closest the two bodies have ever been).

mod_df <- df[35:60, 1:11]


knitr::kable(head(mod_df), format = "html") %>% 
  kable_styling(bootstrap_options = c("striped", "condensed"))

par(mfrow = c(3, 3))
set.seed(23)

hist(mod_df$Leg.Succ.to.Fail.Ratio)
hist(mod_df$Leg.Succ)
hist(mod_df$Leg.Fail)


ratio_means <- rep(mod_df$Leg.Succ.to.Fail.Ratio, 50)

for(i in mod_df$Leg.Succ.to.Fail.Ratio){
   rat_means <- sample(mod_df$Leg.Succ.to.Fail.Ratio, 15, replace = FALSE)
   ratio_means[i] <- mean(rat_means)
   }

hist(rat_means)

success_means <- rep(mod_df$Leg.Fail, 50)

for(i in mod_df$Leg.Succ){
   s_means <- sample(mod_df$Leg.Succ, 15, replace = FALSE)
   success_means[i] <- mean(s_means)
   }

hist(s_means)

fail_means <- rep(mod_df$Leg.Fail, 50)

for(i in mod_df$Leg.Fail){
   f_means <- sample(mod_df$Leg.Fail, 15, replace = FALSE)
   fail_means[i] <- mean(f_means)
   }

hist(f_means)

par(mfrow = c(2, 2))
set.seed(23)

hist(mod_df$Cong.Dim1)
hist(mod_df$Cong.Dim2)


dim1_means <- rep(mod_df$Cong.Dim1, 50)

for(i in mod_df$Cong.Dim1){
   d1_means <- sample(mod_df$Cong.Dim1, 15, replace = FALSE)
   dim1_means[i] <- mean(d1_means)
   }

hist(d1_means)

dim2_means <- rep(mod_df$Cong.Dim2, 50)

for(i in mod_df$Cong.Dim2){
   d2_means <- sample(mod_df$Cong.Dim2, 15, replace = FALSE)
   dim2_means[i] <- mean(d2_means)
   }

hist(d2_means)

plot(mod_df$Cong.Dim1, mod_df$Cong.Dim2)

a <- list(
  autotick = FALSE,
  ticks = "outside",
  tick0 = 1961,
  dtick = 4
  )

yt <- list(title = " ")

suctrend <- plot_ly(mod_df, x = ~year, y = ~Leg.Fail, name = "Number of Legislative Failures", type = 'scatter', mode = 'lines') %>%
  add_trace(y = ~Leg.Succ, name = "Number of Legislative Successes", mode = 'lines+markers') %>% 
  layout(yaxis = yt, hovermode = 'compare')

suctrend

rattrend <- plot_ly(mod_df, x = ~year, y = ~Leg.Succ.to.Fail.Ratio, name = "Legislative Success to Failure Ratio", type = 'scatter', mode = 'lines') 

rattrend

Have there been significant differences in the amount of presidential support from the House and Senate?

Unfortunately the data we have don’t really allow us to test this. The distribution for the House PSS was left-skewing and even if we accept that 67 observations are enough to move forward those observations really aren’t independent since individuals are likely to appear in more than one year’s House or Senate.

Is there more successful legislative activity when Congress supports the President?

First let’s look at the summary data for successful legislative activity and for Presidential Support.

summary(mod_df$Leg.Succ)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   148.0   480.8   544.0   560.5   634.5   983.0

sd(mod_df$Leg.Succ)

## [1] 162.0038

summary(mod_df$Avg.PSS)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   46.10   54.33   59.15   59.34   65.28   70.20

sd(mod_df$Avg.PSS)

## [1] 7.241938

cor(mod_df$Leg.Succ, mod_df$Avg.PSS)

## [1] -0.03354053

plot(mod_df$Leg.Succ, mod_df$Avg.PSS)

There is very little indication that the to metrics are related.

Is there more successful legislative activity when Congress leans more to the right or to the left?

Summary of NOMINATE Dimension 1

summary(mod_df$Cong.Dim1)

##      Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
## -0.071406 -0.021727  0.009786  0.011494  0.043559  0.093658

sd(mod_df$Cong.Dim1)

## [1] 0.04777741

cor(mod_df$Leg.Succ, mod_df$Cong.Dim1)

## [1] 0.3008356

plot(mod_df$Leg.Succ, mod_df$Cong.Dim1)

Summary of NOMINATE Dimension 2

summary(mod_df$Cong.Dim2)

##     Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
## -0.02836  0.00834  0.02327  0.04323  0.08367  0.16430

sd(mod_df$Cong.Dim2)

## [1] 0.05022251

cor(mod_df$Leg.Succ, mod_df$Cong.Dim2)

## [1] 0.5013425

plot(mod_df$Leg.Succ, mod_df$Cong.Dim2)

There doesn’t appear to be any strong connection between legislative success and NOMINATE scores.

Is there more legislative success under a Republican or Democratic President

H₀ = The President’s Party has no influence on the legislative success ratio.
H_A = The President’s Party has an influence on the legislative success ratio.

First we need to isolate the legislative successes by President’s Party

p <- c("Party", "Leg.Succ.to.Fail.Ratio")

pp_rat <- data.frame("Party" = c(mod_df$Party), "Leg.Succ.to.Fail.Ratio" = c(mod_df$Leg.Succ.to.Fail.Ratio))

hist(pp_rat$Leg.Succ.to.Fail.Ratio)

There are only 26 records and the distribution is not clearly normally distributed, but we’re going to move forward anyway.

ppr_gop <- subset(pp_rat, Party == "Republican ")
ppr_dem <- subset(pp_rat, Party == "Democratic ")

ppr_gop

##          Party Leg.Succ.to.Fail.Ratio
## 1  Republican                2.960000
## 2  Republican                2.246575
## 3  Republican                2.142105
## 4  Republican                2.014019
## 13 Republican                2.077519
## 14 Republican                2.733333
## 15 Republican                1.574257
## 16 Republican                2.037500
## 17 Republican                1.693548
## 18 Republican                1.851449
## 19 Republican                1.686106
## 20 Republican                3.291262
## 21 Republican                2.297030
## 22 Republican                3.628866
## 23 Republican                1.115242
## 24 Republican                1.159705
## 25 Republican                1.830189
## 26 Republican                2.637860

ppr_dem

##          Party Leg.Succ.to.Fail.Ratio
## 5  Democratic                2.169291
## 6  Democratic                1.889328
## 7  Democratic                1.214418
## 8  Democratic                1.563492
## 9  Democratic                1.721154
## 10 Democratic                1.782443
## 11 Democratic                1.975352
## 12 Democratic                2.131579

pp_mtx <- data.frame("Party" = c("Democrat", "Republican"), "Mean.Ratio" = c(mean(ppr_dem$Leg.Succ.to.Fail.Ratio), mean(ppr_gop$Leg.Succ.to.Fail.Ratio)), "SD.Ratio" = c(sd(ppr_dem$Leg.Succ.to.Fail.Ratio), sd(ppr_gop$Leg.Succ.to.Fail.Ratio)))

pp_mtx

##        Party Mean.Ratio  SD.Ratio
## 1   Democrat   1.805882 0.3139926
## 2 Republican   2.165365 0.6767097

Checking for 95% confidence

Difference in means = 2.165 - 1.81 = 0.335

n = 26

Df = 25

s1 = SD of Democratic = 0.314

s2 = SD of Republican = 0.677

A = \(\frac{s1^2}{n}\) = 0.004

B = \(\frac{s2^2}{n}\) = 0.018

SE = \(\sqrt{A+B}\) = 0.148

t*SE for 95% CI= 2.06

0.335 \(\pm\) 2.06 x 0.148 \(\Rightarrow\) CI = (-0.029, 0.641)

T = \(\frac{0.335}{0.306}\) = 1.09 \(\Rightarrow\) p > 0.10

As the p-value is more than the \(\alpha\) and as the CI passes through 0 we cannot reject the null hypothesis. It is possible that the President’s political party influences the amount of successful legislative action.