notebook.community

Edit and run

Team Colonials

Final Project Presentation

Awilda Lopez, Bala Venkatesan, Faith Bradley, Rebecca Bilbro

Cohort 2, Fall 2014

The Problem:

In the first three decades after OSHA was created in 1971, workplace fatalities dropped more than 65%, even as US employment doubled. Now worker deaths are down from about 38 a day in 1970 to 12 a day in 2013. But we've hit a plateau. OSHA's budget is small and not likely to increase. The agency can only do so much outreach, training, enforcement, and standard-setting. How can the agency make the most of its limited resources?

The Goal:

Use OSHA data to build a tool that will help the agency develop targeting schemas for outreach, training, enforcement, and regulation.

The Hypotheses:

1. Variation by state or area - Maybe fewer workers are being killed in State X while more workers are being killed in State Y? Or certain field offices are more overwhelmed than others?
2. Impact of the recession - Are there fewer fatalities in construction because of the housing market collapse? Has the recession depressed the number of worker fatalities, artificially diminishing the impact of hazardous workplaces?
3. Time of year - Are many of the most hazardous jobs seasonal?
4. Industries are changing - As hazards decrease in some traditional industries (e.g. traditional manufacturing), are more workers being killed in new industries (e.g. green energy)?

The Data:

The primary data sources are the OSHA website (http://www.osha.gov) and the Department of Labor Enforcement Data Catalog (http://ogesdw.dol.gov/).

Understanding Workplace Fatality/Catastrophe Data





In [1]:



    


import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import statsmodels.formula.api as sm
from sklearn.linear_model import LinearRegression
from scipy import stats
%matplotlib inline
What does the dataset look like?




In [2]:



    


oshadata = pd.read_csv("incidents_by_year.csv", index_col="event_year")
oshadata[:5]



    


















    






/Library/Python/2.7/site-packages/pandas/io/parsers.py:1154: DtypeWarning: Columns (4) have mixed types. Specify dtype option on import or set low_memory=False.
  data = self._reader.read(nrows)










    
Out[2]:










  
    

      
      area_office
      state
      fatality
      industry_code
      description
      keywords
      summary_nr
    

    

      event_year
      
      
      
      
      
      
      
    

  
  
    

      1972
       854910
                UTAH
        NaN
       7359
       Employee suffers partial thumb amputation in t...
          SLIP,TABLE SAW,AMPUTATED,THUMB,GUARD,BLADE,SAW
       170683817
    

    

      1973
       524700
                OHIO
       True
       3444
          Employee fatally crushed by falling automobile
       CRUSHED,STRUCK BY,FALLING OBJECT,AUTOMOBILE,BO...
        14514202
    

    

      1973
       316400
       WEST VIRGINIA
       True
       3334
         Employee killed while dismantling crane gearbox
       DISMANTLING,CRANE,STRUCK BY,FALL,CRUSHED,UNTRA...
        14481709
    

    

      1974
       627400
               TEXAS
       True
       1381
       One employee killed, two injured when struck b...
       OIL WELL DRILLING,TONGS,STRUCK BY,HAND TOOL,FL...
        14413066
    

    

      1974
       627400
               TEXAS
       True
       1381
                   Employee killed in trailer house fire
       FIRE,BURN,SMOKING,FLAMMABLE VAPORS,OIL WELL DR...
        14413074
Yeah, but what does it LOOK like? Let's check out a visualization of these incidents as a time series.




In [3]:



    


df = pd.DataFrame.from_csv('incident_totals_year.csv', parse_dates=False)
df.osha_incidents.plot(color='r',lw=1.3)
plt.title('Workplace Fatalities and Catastrophes by Year')
plt.xlabel("1983-2012")
plt.ylabel("Fatal/Catastrophic Incidents")
plt.show()
It turns out that, because of irregularities in the way that these reports were logged before 1990, we should ignore what appears to be a trend in volume from 1983-1990. We know from the Bureau of Labor Statistics that there were actually more like 10,000 fatalities per year during that period. The graph also suggests that fatal events trail off after 2009, but it turns our that this is due to data maturation issues. BLS reports show that the number of fatal workplace incidents has been fairly steady at around 4,500 per year since 2010.
So let's do a histogram to see the frequency of events by year, but we'll just focus in on 1990-2010




In [4]:



    


subset = df.loc[1990:2010]
subset.plot(kind='bar', stacked=False)
plt.title("Frequency of fatalities and catastrophic events by year")
plt.xlabel("1990-2010")
plt.ylabel("Frequency")
plt.show()

The Analysis:

Hypothesis One (Rebecca): Understanding Regional Variations in Workplace Fatality/Catastrophe Data
One of our hypotheses about the data is that there may be significant regional variation in the incidence of workplace fatalities and catastrophes. By getting a sense of where the most workers are dying by location, we could assist OSHA in directing more staff and resources to the most dangerous parts of the country.
To test this hypothesis, we can analyze our data in several ways. We can look at the frequency of events by state, or at a more granular level, by area office.
Let's start by looking at variations by state




In [5]:



    


statedata = pd.read_csv("incidents_state_totals.csv")
statedata



    


















    
Out[5]:










  
    

      
      state
      totals
    

  
  
    

      0 
                    ALABAMA
        1115
    

    

      1 
                     ALASKA
         446
    

    

      2 
                    ARIZONA
        1307
    

    

      3 
                   ARKANSAS
         722
    

    

      4 
                 CALIFORNIA
       38477
    

    

      5 
                   COLORADO
         848
    

    

      6 
                CONNECTICUT
        1065
    

    

      7 
                   DELAWARE
         125
    

    

      8 
       DISTRICT OF COLUMBIA
          33
    

    

      9 
                    FLORIDA
        3225
    

    

      10
                    GEORGIA
        1828
    

    

      11
                     HAWAII
         395
    

    

      12
                      IDAHO
         417
    

    

      13
                   ILLINOIS
        2553
    

    

      14
                    INDIANA
        1724
    

    

      15
                       IOWA
        1779
    

    

      16
                     KANSAS
        1396
    

    

      17
                   KENTUCKY
        1029
    

    

      18
                  LOUISIANA
        1257
    

    

      19
                      MAINE
         274
    

    

      20
                   MARYLAND
        3908
    

    

      21
              MASSACHUSETTS
        1491
    

    

      22
                   MICHIGAN
        1485
    

    

      23
                  MINNESOTA
         885
    

    

      24
                MISSISSIPPI
         690
    

    

      25
                   MISSOURI
         879
    

    

      26
                    MONTANA
         273
    

    

      27
                   NEBRASKA
         615
    

    

      28
                     NEVADA
        1042
    

    

      29
              NEW HAMPSHIRE
         386
    

    

      30
                 NEW JERSEY
        1754
    

    

      31
                 NEW MEXICO
         350
    

    

      32
                   NEW YORK
        2928
    

    

      33
             NORTH CAROLINA
        2833
    

    

      34
               NORTH DAKOTA
         421
    

    

      35
                       OHIO
        2630
    

    

      36
                   OKLAHOMA
         768
    

    

      37
                     OREGON
        4052
    

    

      38
               PENNSYLVANIA
        1961
    

    

      39
               RHODE ISLAND
         156
    

    

      40
             SOUTH CAROLINA
        1591
    

    

      41
               SOUTH DAKOTA
           0
    

    

      42
                  TENNESSEE
        1268
    

    

      43
                      TEXAS
        4728
    

    

      44
                       UTAH
        1580
    

    

      45
                    VERMONT
         159
    

    

      46
                   VIRGINIA
        2402
    

    

      47
                 WASHINGTON
        2173
    

    

      48
              WEST VIRGINIA
         440
    

    

      49
                  WISCONSIN
        1167
    

    

      50
                    WYOMING
         276
    

  




















In [6]:



    


statedata.plot(kind='bar', stacked=False)
plt.title("Frequency of fatalities and catastrophic events by state")
plt.xlabel("United States")
plt.ylabel("Frequency")
plt.show()
So California is the clear standout in terms of sheer volume. Why is that? California is a state plan state, meaning that it is not under the jurisdiction of Federal OSHA, but that alone doesn't really explain why they would have so many more fatalities (or reports of fatalities).
And hmmm...something strange is definitely going on in South Dakota. Let's try to figure out why those reports are missing from this dataset.
In any case, let's look a little deeper into the areas within the states where the most workers are dying on the job.
Hypothesis Two (Bala): Examining the Relationship Between the Housing Market Collapse and Workplace Fatality/Catastrophe Data
Hypothesis Three (Awilda): Investigating the Seasonality of Workplace Fatalities and Catastrophes
Hypothesis Four (Faith): Exploring the Changing Industrial Landscape in Relation to Workplace Fatalities and Catastrophes

The Interpretations:

(to be added later)

The Conclusions:

(to be added later)




In [6]:

Content source: rebeccabilbro/capstone

Similar notebooks:

notebook.community | gallery | about