Senate Votes ETL and Database Population

By Nigel Story

Introduction

This is the ETL notebook that populates my locally hosted senate MySQL database, used for the machine learning notebook Analysis of U.S. Senate Polarization. Here, we scrape publicly available voting records for each U.S. senator, along with data about the proceedings and issues at vote.

Packages

In [2]:
import requests
from bs4 import BeautifulSoup
import xml.etree.ElementTree as ET
import pandas as pd
import numpy as np
import re
import os
from io import StringIO
from tqdm import tqdm
from helpers.db_connection import DBConnect

Proceedings ETL

We can extract data for the proceedings from senate.gov using a BASH script and the wget command. This script can be found in "./bash_scripts/get_data.bash" within this repository.

In [1]:
#!mkdir /Users/nastory/repos/senate_votes/xml_data
#!cd /Users/nastory/repos/senate_votes/xml_data
#!bash ../bash_scripts/get_data.bash

After retrieving the data as xml, we will need to convert it to a pandas data frame. This will make it easier to import into the database.

In [3]:
xml_path = './senate_votes/xml_data/'
vote_roll_call_url = 'https://www.senate.gov/legislative/LIS/roll_call_lists/roll_call_vote_cfm.cfm?congress={}&session={}&vote={}'
In [4]:
data = []
files = os.listdir(xml_path)
files.sort()
xml_check = re.compile('\.xml')
In [5]:
def check_value(element_name, parent):
    # coalesce missing values with empty string
    check = parent.find(element_name)
    if check is None:
        return ''
    else:
        return check.text
In [6]:
for f in files:
    if xml_check.search(f):
        tree = ET.parse(xml_path + f)
        root = tree.getroot()
        
        congress = root[0].text
        session = root[1].text
        year = root[2].text
        
        for vote in root.iter('vote'):
            data.append([
                congress, 
                session, 
                year, 
                vote.find('vote_number').text,
                vote.find('vote_date').text,
                check_value('issue', vote),
                check_value('question', vote),
                vote.find('title').text,
                vote.find('vote_tally').find('yeas').text,
                vote.find('vote_tally').find('nays').text,
                check_value('result', vote)
            ])
        
    else:
        pass
In [7]:
column_names = ['congress', 'session', 'year', 'vote_number', 'vote_date', 'issue',
                'question', 'title', 'yeas', 'nays', 'result']
df = pd.DataFrame(data, columns=column_names)
In [8]:
num_cols = ['congress', 'session', 'year', 'yeas', 'nays']
df[num_cols] = df[num_cols].astype(float)
df[num_cols] = df[num_cols].fillna(0)
In [9]:
df['vote_date'] = df['vote_date'].astype(str) + '-' + df['year'].astype(str)
df['vote_date'] = pd.to_datetime(df['vote_date'], format='%d-%b-%Y.0')
In [10]:
df['question'] = df['question'].str.replace(r'\n\s*', '')
In [11]:
with DBConnect('senate', autocommit=True) as cnx:
    cnx.df2db(df, 'proceedings', method='insert', if_dup_key_replace=True)

Senators' Votes ETL

I attempted two methods of retrieving senators' voting records. The first was to scrape the data from senate.gov, but the volume of requests to the server that I was making was too high, so I was temporarily banned from the site. The alternative was a site called govtrack.us, which had the data in a more accessable format to scrape anyway, so it all worked out.

Method 1

senate.gov banned me, so the below won't work anymore...

In [12]:
# def html2votes(response_text, congress, session, vote_number):
#     soup = BeautifulSoup(response_text)
#     votes = soup.find_all('div', {'class': 'newspaperDisplay_3column'})[0]

#     vote_text = re.sub(r'<[\s\S]*>', '', votes.text)
#     vote_text = vote_text.replace(' (', ',').replace('), ', ',').replace('-', ',')
#     vote_text = 'senator,party,state,vote\n' + vote_text

#     out = pd.read_csv(StringIO(vote_text), delimiter=',')
#     out['congress'] = congress
#     out['session'] = session
#     out['vote_number'] = vote_number
    
#     cols = ['congress', 'session', 'vote_number', 'senator', 'party', 'state', 'vote']
#     out = out[cols]
    
#     return out
In [13]:
# failed = []
# df2 = pd.DataFrame()
# for row in params:
#     url = vote_roll_call_url.format(int(row[0]), int(row[1]), row[2])
#     res = requests.get(url)
    
#     if res.status_code != 200:
#         failed.append([url, res.status_code])
        
#     else:
#         df2 = pd.concat([df2, html2votes(res.text, int(row[0]), int(row[1]), row[2])], axis=0)
        

Method 2

In [14]:
gt_url = 'https://www.govtrack.us/congress/votes/{}-{}/s{}/export/csv'
params = df[['congress', 'session', 'vote_number', 'year']].to_numpy()

Download data

In [15]:
# df2 = pd.DataFrame()
# failed = []
# for row in params:
#     res = requests.get(gt_url.format(int(row[0]), int(row[3]), int(row[2])))
#     if res.status_code == 200:
#         with open('./csv_data/{}-{}-{}.txt'.format(int(row[0]), int(row[3]), int(row[2])), 'w') as f:
#             f.write(res.text)
#     else:
#         failed.append([gt_url.format(int(row[0]), int(row[3]), int(row[2])), res.status_code])
In [16]:
#len(failed)

Read files and import to db.

In [37]:
path = './csv_data/'
regex = re.compile(r'person,state,district,vote,name,party[\s\S]*')
failed = []
senator_set = set()

with DBConnect('senate', autocommit=True) as cnx:
    for file in os.listdir(path):
        try:
            c, y, v = file.split('-')
            with open(path + file, 'r') as f:
                txt = f.read()
                match = regex.search(txt)
                if match:
                    data_txt = match.group(0)
                    
            dummy = pd.read_csv(StringIO(data_txt))
            dummy['name'] = dummy['name'].str.replace(r'\s\[[\S\s]*\]', '')

            tups = [tuple(x) for x in dummy[['person', 'name', 'state', 'district', 'party']].to_numpy()]
            for t in tups:
                senator_set.add(t)

            dummy['congress'] = c
            dummy['year'] = y
            dummy['vote_number'] = v.replace('.txt', '').zfill(5)
            dummy = dummy[['congress', 'year', 'vote_number', 'person', 'vote']]
            dummy.rename(columns={'person': 'senator_id'}, inplace=True)

            cnx.df2db(dummy, 'votes', method='insert', if_dup_key_replace=True)
            
        except Exception as e:
            failed.append([file, str(e)])
    
In [47]:
len(failed)
Out[47]:
0
In [55]:
sen_df = pd.DataFrame(senator_ls, columns=['senator_id', 'name', 'state', 'district', 'party'])
In [45]:
sen_df['name'] = sen_df['name'].str.strip()
In [46]:
with DBConnect('senate', autocommit=True) as cnx:
    cnx.df2db(sen_df, 'senators', method='insert', if_dup_key_replace=True)

Test

Now that the data is in the database, we'll perform a few test queries.

In [53]:
with DBConnect('senate') as cnx:
    proceedings = pd.read_sql(con=cnx.cnx, sql='select * from proceedings limit 100')
    votes = pd.read_sql(con=cnx.cnx, sql='select * from votes limit 100')
    senators = pd.read_sql(con=cnx.cnx, sql='select * from senators limit 100')
In [50]:
proceedings.head()
Out[50]:
congress year vote_number session vote_date issue question title yeas nays result
0 101 1989 00001 1 1989-01-25 PN128 On the Nomination Nomination - Baker 99 0 Confirmed
1 101 1989 00002 1 1989-01-25 PN133 On the Nomination Nomination - Dole 99 0 Confirmed
2 101 1989 00003 1 1989-01-25 PN139 On the Nomination Nomination - Darman 99 0 Confirmed
3 101 1989 00004 1 1989-01-31 PN140 On the Nomination Nomination - Hills 100 0 Confirmed
4 101 1989 00005 1 1989-01-31 PN132 On the Nomination Nomination - Mosbacher 100 0 Confirmed
In [51]:
votes.head()
Out[51]:
congress year vote_number senator_id vote
0 101 1989 00001 300005 Yea
1 101 1989 00001 300008 Yea
2 101 1989 00001 300009 Yea
3 101 1989 00001 300010 Yea
4 101 1989 00001 300012 Yea
In [54]:
senators.head()
Out[54]:
senator_id name state district party
0 300001 Sen. Daniel Akaka HI None Democrat
1 300002 Sen. Lamar Alexander TN None Republican
2 300003 Sen. Wayne Allard CO None Republican
3 300004 Sen. George Allen VA None Republican
4 300005 Sen. Max Baucus MT None Democrat
In [ ]: