Scribd Logo
Upload

Welcome to Scribd!

  • 222 views

    Benchmark Specification Sheet

    Uploaded by

    Anthony Reed
    This document provides details on the data and methodology used in the Benchmark election prediction model. The model uses open source data such as census information, generic ballot polls, and historical election results as well as some proprietary data. It analyzes districts nested within states and outputs predicted margins that classify races as tossup, lean, or likely. Local polling is incorporated by averaging results with model predictions. The model does not account for candidate quality factors.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd

    Benchmark Specification Sheet

    Uploaded by

    Anthony Reed
    222 views2 pages
    This document provides details on the data and methodology used in the Benchmark election prediction model. The model uses open source data such as census information, generic ballot polls, and historical election results as well as some proprietary data. It analyzes districts nested within states and outputs predicted margins that classify races as tossup, lean, or likely. Local polling is incorporated by averaging results with model predictions. The model does not account for candidate quality factors.

    Original Description:

    Specification sheet for the Benchmark Politics model

    Copyright

    © © All Rights Reserved

    Available Formats

    DOCX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document provides details on the data and methodology used in the Benchmark election prediction model. The model uses open source data such as census information, generic ballot polls, and historical election results as well as some proprietary data. It analyzes districts nested within states and outputs predicted margins that classify races as tossup, lean, or likely. Local polling is incorporated by averaging results with model predictions. The model does not account for candidate quality factors.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Download as docx, pdf, or txt
    222 views2 pages

    Benchmark Specification Sheet

    Uploaded by

    Anthony Reed
    This document provides details on the data and methodology used in the Benchmark election prediction model. The model uses open source data such as census information, generic ballot polls, and historical election results as well as some proprietary data. It analyzes districts nested within states and outputs predicted margins that classify races as tossup, lean, or likely. Local polling is incorporated by averaging results with model predictions. The model does not account for candidate quality factors.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Download as docx, pdf, or txt
    You are on page 1of 2

    Benchmark Model Specification Sheet

    Data Used
    The Benchmark model uses a variety of open source data and some closed, proprietary
    data to make its predictions. Among the open source data used are:

    - Census data, obtained from IPUMS: https://usa.ipums.org/usa/


    o How we use it: To obtain demographic variables, such as race, income,
    population, age distribution, and economic data on industries.

    - National Generic Ballot, obtained from Fivethirtyeight:


    https://projects.fivethirtyeight.com/congress-generic-ballot-polls/
    o How we use it: To adjust the national environment for all of the races in the
    model.

    - Historical election data for districts, obtained from Dailykos:


    https://www.dailykos.com/stories/2012/11/19/1163009/-Daily-Kos-Elections-
    presidential-results-by-congressional-district-for-the-2012-2008-elections
    o How we use it: To determine the presidential voting history of a district for
    the model, and establish how it differs from the congressional voting history.

    - Incumbency and congressional election data, obtained from Ballotpedia:


    https://ballotpedia.org
    o How we use it: To handicap incumbents properly and to see if they tend to
    over-perform or under-perform the presidential margins in their district.

    - Cook PVI values, obtained from The Cook Report:


    http://cookpolitical.com/introducing-2017-cook-political-report-partisan-voter-
    index
    o How we use it: To give a standardized lean to each district in the model.

    - Party Affiliation Changes, obtained from proprietary data


    o How we use it: To see how a district is moving in one direction or another in
    their partisan lean.

    - Polling and Trump Approval, obtained from pollster’s websites.


    o How we use it: Sometimes, local polling exists for the races. The model can
    take these polls into account. Trump approval is also useful at the state level.
    Methodology

    All data is collected and arranged in an HLM format – districts nested in states. This
    adequately takes into account the differing political landscape that is present in each state
    according to statewide variables. Each district also has its own set of variables that are
    controlled for when making a prediction.

    Ultimately, running the model in STATA will yield a predicted margin for the Republican
    candidate. If that margin is negative, then the Democrat is favored to win by that amount. If the
    margin is positive, the Republican is favored to win. We classify our results into general
    categories, such as Tossup (0-3%), Lean (3-6%), Likely (6-12%).

    A special word on local polling: Sometimes, local polling is done on house races. These
    need to be incorporated into the model. Having a dummy variable inserted into the model for
    polling doesn’t work well for races where there is no polling at all.

    The best way to do this, I have found, is to treat polling as a separate event and average
    the average polling results with our modeling. Generally, they are not terribly far apart.
    However, in cases where they are far apart, we feel like this allows us to project if the polling is
    overestimating a candidate or not. We weight this average based on the quality of the polling.

    It is important to note that there are some things that the model does not take into
    account. Most notably, candidate quality is not something we are modeling.

    Local polling will mostly capture candidate quality, so in cases where local polling exists,
    this is not so much a problem. In races with no local polling and little national coverage, we do
    not feel it is prudent to manually decide for ourselves what is a quality candidate and what is
    not, so we report the model numbers as is, regardless of the quality of the candidate.

    You might also like