An Interactive Visual Analytics Dashboard for the Employment Situation Report

An Interactive Visual Analytics Dashboard for the Employment Situation Report

Benjamin Bengfort1, Xintong Han2, Assaf Magen3, and Hao Zhou4

Department of Computer Science, University of Maryland {bengfort1,hzhou4}@cs.umd.edu, {xintong2,amagen3}@umd.edu

May 12, 2015

The Employment Situation Report

“Its information is widely anticipated, forecasted and used by Wall Street firms, their economists and many business decision-makers. It may even impact broader public and

corporate confidence, and therefore future business and hiring decisions.”

What You Need To Know About the Employment Report

New York Times Reporting on the Jobs Report

The visual analytics process combines automatic and visual analysis methods with a tight coupling through human interaction in order to gain knowledge from data.

Visual Analytics

D. A. Keim, J. Kohlhammer, G. Ellis, and F. Mansmann, Mastering The Information Age-Solving Problems with Visual Analytics. Florian Mansmann, 2010.

Visualization

Knowledge

Models

Data

User Interaction

Model Viz

Model Building

Mapping

Transformation

Data Mining

Feedback Loop

The Sensemaking Process

J. Heer and B. Shneiderman, “Interactive dynamics for visual analysis,” Queue, vol. 10, no. 2, p. 30, 2012.

High Level Overview Zoom & Filter Details on

Demand

Requirements for BLS Analysis

- Interactive, hypothesis-driven exploration of multiple data sets that can be verified and exported.

- General accessibility without prerequisite domain knowledge.

- Rapid and correct at-a-glance insights.

Requirements for BLS Analysis

- Interactive, hypothesis-driven exploration of multiple data sets that can be verified and exported.

- General accessibility without prerequisite domain knowledge.

- Rapid and correct at-a-glance insights.

Requirements for BLS Analysis

- Interactive, hypothesis-driven exploration of multiple data sets that can be verified and exported.

- General accessibility without prerequisite domain knowledge.

- Rapid and correct at-a-glance insights.

The Jobs Report is based on the following, national surveys:- Current Population Survey (CPS): surveys households

and relates demographic trends to employment and labor force characteristics.

- Current Employment Statistics (CES): surveys employers for industry-specific details concerning employment.

Additionally, for Census region specific information:- Local Area Unemployment (LAU): household survey

that relates state and metro regions with demographic and labor force dimensions.

Available Data Sets

The Jobs Report is based on the following, national surveys:- Current Population Survey (CPS): surveys households

and relates demographic trends to employment and labor force characteristics.

- Current Employment Statistics (CES): surveys employers for industry-specific details concerning employment.

Additionally, for Census region specific information:- Local Area Unemployment (LAU): household survey

that relates state and metro regions with demographic and labor force dimensions.

Available Data Sets

The Jobs Report is based on the following, national surveys:- Current Population Survey (CPS): surveys households

and relates demographic trends to employment and labor force characteristics.

- Current Employment Statistics (CES): surveys employers for industry-specific details concerning employment.

Additionally, for Census region specific information:- Local Area Unemployment (LAU): household survey

that relates state and metro regions with demographic and labor force dimensions.

Available Data Sets

The BLS Pipeline

T. Ojeda, S. P. Murphy, B. Bengfort, and A. Dasgupta, Practical Data Science Cookbook. Packt Publishing Ltd, 2014.

Data Ingestion Data Wrangling Data Storage

APIFront-End Application

OverviewZoom & Filter

Details on Demand

Data Ingestion

- Data challenge I: management of- 1,684 time series ids to fetch from BLS- 305,445 records from Jan 2000 - Feb 2015

- Data challenge II: monthly refresh of data- Solution: lightweight Python wrapper using the

BLS API version 2.0 (with API key authentication).

Data Ingestion

- Data challenge I: management of- 1,684 time series ids to fetch from BLS- 305,445 records from Jan 2000 - Feb 2015

- Data challenge II: monthly refresh of data- Solution: lightweight Python wrapper using the

BLS API version 2.0 (with API key authentication).

Data Ingestion

- Data challenge I: management of- 1,684 time series ids to fetch from BLS- 305,445 records from Jan 2000 - Feb 2015

- Data challenge II: monthly refresh of data- Solution: lightweight Python wrapper using the

BLS API version 2.0 (with API key authentication).

import blsapiimport prettytable

if __name__ == '__main__':

## Demo Series series = ['LNS12000000', 'LNS13000000', 'LNS10000000'] result = blsapi.bls_series(series, startyear='2010', endyear='2015')

## Fields fields = ["series id", "year", "period", "value", "footnotes"]

## Pretty print a table of the results for s in result['Results']['series']: table = prettytable.PrettyTable() for item in s['data']: table.add_row([item[k] for k in fields])

print table.get_string()

Data Wrangling

- Data challenge 0: Initial BLS IDs?- Data challenge III: BLS APIs come with no meta

data associated with them. Data must be parsed directly from the description text.

- Data challenge IV: Integration of other data sources and general computation.

Data Wrangling

- Data challenge 0: Initial BLS IDs?- Data challenge III: BLS APIs come with no meta

data associated with them. Data must be parsed directly from the description text.

- Data challenge IV: Integration of other data sources and general computation.

Data Wrangling

- Data challenge 0: Initial BLS IDs?- Data challenge III: BLS APIs come with no meta

data associated with them. Data must be parsed directly from the description text.

- Data challenge IV: Integration of other data sources and general computation.

Web Application and API

Application Architecture

BLS API PostgreSQLDatastore

Ingestion & Wrangling

Multiple Dashboard Interface

On to the Song & Dance(Demonstration)

Usability Testing

- 8 participants over 2 studies: 5 research driven tasks and 8 questions.

- Users were asked to interact with the application and “think aloud” while researching their tasks. They completed a usability questionnaire on completion.

- Results from the study and feedback were incorporated into a prioritized change list.

Future Work

- Automatic headline generation to match the time series that are added to the explorer.

- Better rate of change handling (specify deltas) or compute differentials across wider ranges of time.

- Enhance the choropleth with more details on demand for local data.

- Automated ingestion mechanism.

Questions?

http://bit.ly/elmr-video

http://bit.ly/elmr-repohttp://elmr.herokuapp.com/

http://bit.ly/elmr-slides