Fuqua - Case: 2008 Democratic Primaries - Clinton vs. Obama
2008 Democratic Primaries – Clinton vs. Obamai Background
It was February 19, 2008. One week earlier, Barack Obama had taken the lead in the delegate count during the Democratic Party’s presidential primaries, the winner of which would face the Republican Party’s nominee in the general election to become the next president of the United States (POTUS).
On that day in February, Hillary Clinton, Obama’s primary opponent, began running ads in Ohio aimed at middle-class, blue-collar voters. One ad, “Night Shift,” closed showing Clinton at her desk: “She understands. She’s worked the night shift, too.” But had Clinton ever worked the night shift? Her spokesperson said it was a “rhetorical reference” to working late nights as a lawyer, First Lady, and senator [1].
Clinton was not alone in her awkward appeals to voters in key demographics. Months earlier at a campaign stop in Iowa, Obama noted that while produce prices had risen in grocery stores, farmers had not benefited from increases in crop prices: “Anybody gone into Whole Foods lately and see what they charge for arugula? I mean, they’re charging a lot of money for this stuff.” At the time, there wasn’t a single Whole Foods in the state of Iowa [2].
What did these missteps say about the candidates’ campaign strategies? Were they targeting the right voters? Had they crafted the right messages for these segments? One key in answering these questions was to analyze votes that had already been cast.
Demographic Data
In November 2007, the U.S. Census Bureau issued its County and City Data Book [3] which contained extensive demographic information by state, county, and city. The Census Bureau grouped the 50 U.S. states into four regions (see Exhibit 1). Within the United States, there were a total of 3,141 counties.
On its website, the Census Bureau released the 2007 data tables from the County and City Data Book. These tables contained information commonly used by marketers to segment a population—gender, age, race, ethnicity, education, income, employment status, knowledge of languages, government dependency, disabilities, home ownership, mobility, population, population density, and region [4]. Key demographic data by county were extracted from these tables and placed in a spreadsheet file.
Vote Data
Candidates for the Democratic nomination won delegates—both pledged delegates and superdelegates—through a complicated process followed closely by many news organizations and depicted in many different ways.1 On its website, CNN displayed an interactive graphic where candidates were depicted as donkeys in a foot race; visitors could roll a cursor over a candidate’s donkeys and the current delegate total would appear [5].
Politicians cared about votes the way marketers cared about sales. Because voting results dictated delegate commitment, major news outlets carefully tracked totals by county. A visitor had access to county vote data (also included in the spreadsheet) by rolling a cursor over a state’s map.2
As of February 19, 2008, of the 2,868 total counties for which county-level vote data would eventually become available, there were 1,131 counties left to report. An estimated 2,490 delegates were already committed, and 2,118 were needed to secure the party’s nomination; there were still 1,744 delegates to be awarded in upcoming states’ primaries and caucuses. (See Exhibit 2 for a list of past and upcoming primaries and caucuses.) Later that evening, the results of Hawaii’s caucus and Wisconsin’s primary would be announced. In these two states, and in over 1,000 counties in other states, it remained to be seen who would vote for Clinton and who would vote for Obama. Whoever won the most votes in these remaining primaries was likely to become the next POTUS.
What Followed and the Role of Analytics in Politics
Indeed, Obama won the Democrat nomination and won two Presidential elections. In the 2008 presidential election, Obama’s targeters had assigned every voter in the country a pair of scores based on the probability that the individual would perform two distinct actions that mattered to the campaign: casting a ballot and supporting Obama.
For each battleground state every week, the campaign’s call centers conducted 5,000 to 10,000 so-called short-form interviews that quickly gauged a voter’s preferences, and 1,000 interviews in a long-form version that was more like a traditional poll. To derive individual-level predictions, algorithms trawled for patterns between these opinions and the data points the campaign had assembled for every voter— as many as one thousand variables each, drawn from voter registration records, consumer data warehouses, and past campaign contacts.
Furthermore, the Obama 2008 campaign also used analytics to increase donations using A/B testing.
1 Pledged delegates were committed to a candidate in proportion to the votes cast for that candidate in the state’s primary or
caucus. Superdelegates (almost 20% of the total number of delegates) were free to commit to any candidate at any time.
2 Several primaries and caucuses were excluded from the county-level vote data set because either Obama’s name did not
appear on the ballot, vote data was not reported by county, or a U.S. county was not represented.
The campaign tried four buttons and six different media (three images and three videos). A full-factorial multivariate test was used to statistically test all the combinations of buttons and media against each other at the same time. Since there were four buttons and six different media that meant a total of 24 (4 x 6) total combinations to test. Every visitor to the splash page was randomly shown one of these combinations and we tracked whether they signed up or not.
The winning variation had a sign-up rate of 11.6%. The original page had a sign-up rate of 8.26%. That’s an improvement of 40.6% in sign-up rate. What does an improvement of 40.6% translate into? Instead of having nearly 10 million people signed up for the campaign, the original page would have obtained only close to 7,120,000 signups. That is a difference of 2,880,000 email addresses. These additional email addresses are essentially responsible for 288,000 more volunteers and an addition $60millions in donations.
This created a new culture, and the Obama 2012 campaign also used data analytics and the experimental method to assemble a winning coalition vote by vote. In doing so, it overturned the long dominance of TV advertising in U.S. politics and created something new in the world: a national campaign run like a local ward election, where the interests of individual voters were known and addressed [6].
Exhibit 1 – Census Regions and Divisions
U.S. Census Bureau’s Regions and Divisions for 2007
Source: County and City Data Book: 2007, 14th ed. (Washington, DC: U.S. Census Bureau, 2007).
Exhibit 2 – Past and Upcoming Democratic Party Primaries and Caucuses
Exhibit 3 – Variable definitions
Data Type Column Headers in the Data sheet Description
Case: 2008 Democratic Primaries – Clinton vs. Obama Instructions
This is a team assignment. Each member of the team receives the same grade. Submission is online. In order to be graded, you need to upload one PDF file (no longer than 5 pages with font size 12pt) and your R script (this should be well commented and run without errors). Any additional material you judge relevant that complements your submission can be submitted as additional files. Make sure that the section number, team number and all names of the team members are clearly listed in the top of the first page of the PDF file. Late submissions (but submitted before in-class discussions) or inappropriately formatted cases will have points deducted. Missed cases are worth 0 points.
Assignment (Important, read carefully)
There are five questions. However, only questions 1 and 2 are mandatory. (That is, you can get full case credit based only on those two questions.)
We will discuss all questions in class. Thus, this is an opportunity for you to work on these tools before class and compare. (Teams submitting responses to Questions 3, 4 and 5 can get bonus/additional points on this case up to 10 points.)
Your answers should be clear and provide unambiguous recommendations when asked. Please provide explanations for your answers and any outputs that you feel are needed to support your argument. Keep in mind that this is an open ended exercise. There is no exact model of reality as discussed in class. You will be evaluated on your modeling of the problem, judgment of which core task to use, appropriateness of the choice of algorithm, and taking all of that to data. (Therefore do not stress about trying to fine tune details.)
An R script is provided with some data cleaning and specific models for Question 4.
Questions
-
Pick two (or more) variables and attempt to show a relation between them via visualization. As discussed before, this requires one to formulate a question, and to communicate clearly a conclusion based on data visualization (specify the why, what, how). (Note that in this question it is not required that the relationship displayed relates to the election.)
-
Provide a model to predict the winning spread of Obama over Clinton measured as percentage of the total vote. Describe clearly the core task, briefly discuss all the models you compared, state which metric is being used to evaluate performance, and how did you chose a final model. Apply and report a K-fold cross validation to evaluate the performance of your chosen model. Based on your final model, predict the winning spread percentages for the test sample (provide the R code that generate your predictions).
Vote Demographic/county data
|
Training set Primaries and caucuses before February 19, 2008 (1,737 rows) |
Available |
Available |
|
Testing set Primaries and caucuses on or after February 19, 2008 (1,131 rows) |
Not Available (submit predictions) |
Available |
-
(Optional Question) In order to explore the data, apply one unsupervised learning tool (e.g., k- means, principal component analysis), interpret and communicate briefly the output (e.g., clusters, latent features), and attempt to obtain insights.
-
(Optional Question) Several sources have been reporting that the demographic composition of the US is changing which can definitely impact how campaigns will be run. In many states, the Hispanic population is growing at a faster pace than others. Looking ahead, provide an estimate for what would have been the average impact on the winning spread for Obama over Clinton (measured in percentage of total voters) had the Hispanic demographic been 5% larger? What if the Black demographic was 5% larger? (Base your response only on the two simple regression models provided in the R file.)
Next, answer the question being careful to isolate the impact of the specific demographic change alone. (This question would be too open ended. In the R starter script we provide a “simple” model with 1771 variables to be used for which we will assume that the Conditional Independence Assumption (CIA) holds.)
5. (Optional Question) Choose one candidate. What kind of advice (based on data analytics) would you provide to your candidate? For example, which voter segment to target with their campaign messages and why? Or, how to allocate resources (budget and volunteer time) across regions and why? How would you communicate such insights?
i This case was created based on the case UVA-QA-0807 which was prepared by Associate Professor Kenneth C. Lichtendahl Jr. and Rohit Gupta (MBA ’13) at Darden School of Business.
