Chapter 44: Capstone 3 — The Campaign Analytics Plan

"A campaign analytics plan is a document you write when you're calm so that you can make good decisions when you're not. On Election Day, nobody's calm. The plan is what keeps you honest." — Nadia Osei, in conversation with her new analytics associate, 60 days out

A Note Before You Begin

The first two capstones asked you to evaluate and audit political data systems — a polling operation, a misinformation tracker — from positions of relative distance. You were the analyst looking in.

This capstone puts you inside the operation.

You are Nadia Osei's new analytics associate on the Garza Senate campaign. You have been hired 60 days before Election Day. Nadia is 31, brilliant, and working 80-hour weeks in a campaign office that smells like pizza boxes and anxiety. She does not have time to explain everything twice. She will give you frameworks, she will give you data, she will give you her honest assessment of what's working and what isn't — and then she will expect you to run with it.

This is what it actually feels like to do political analytics in a contested campaign. It is not comfortable. The data doesn't always tell you what you want to hear. The campaign manager doesn't always agree with what the data shows. Budget is always tighter than you'd like, time is always shorter, and the stakes are real: an election that real people will win or lose, in a state where that outcome will affect real policies affecting real lives.

Build the plan like it matters. Because it does.

Section 1: Introduction — The Analytics Director's Challenge

The Hire

Nadia Osei did not plan to hire a new analytics associate 60 days out. She planned to bring someone on in the spring, train them through the summer, and have them fully integrated by Labor Day. What actually happened: her previous associate took a better-paying job at a tech company in August, citing "fewer existential stakes," and Nadia has been running the analytics operation solo for three weeks by the time you walk through the door.

She meets you at the campaign's field office in the state capital — a rented suite above a tax preparation service, crowded with folding tables, laptops, lawn signs, and the organizational residue of a campaign that has been accelerating for nine months. The primary was competitive. The general has been competitive. Everything has been competitive, and Nadia is tired in the way that people who have been excellent under sustained pressure get tired: competent and efficient on the surface, but running on fumes underneath.

She shakes your hand, hands you a manila folder, and says: "I'll give you ten minutes of context, then you're going to read everything in that folder, and then we're going to talk about what needs to happen before Election Day. Ready?"

You are. Or at least, you say you are.

The folder contains: the most recent Meridian Research Group internal poll (three weeks old), a county-level demographic breakdown from the ODA dataset, a summary of the campaign's current field program status, and a one-page budget overview. It also contains a handwritten note from Nadia that reads: "This is not everything. This is what I can give you right now. Ask me what you need."

The State of the Race

The Garza-Whitfield Senate race is genuinely competitive with 60 days to go. The most recent Meridian internal poll shows Maria Garza at 46 percent, Tom Whitfield at 44 percent, with 10 percent undecided or soft — a statistically significant but narrow lead. The public polling average shows Garza leading by approximately 2.5 points.

The demographics of the state are complex. Garza's coalition depends on strong performance with Latino voters (who make up 32 percent of the electorate), solid margins with Black voters (18 percent), and competitive performance with college-educated white voters in the suburbs — particularly women. Whitfield's coalition is anchored in rural white non-Hispanic voters, working-class voters of all backgrounds who respond to his economic populism, and irregular Republican voters who don't turn out in every cycle but could be mobilized by his insurgent brand.

The electoral geography divides roughly as follows: Garza is running up the score in the two major metro areas, competitive in the suburban ring counties, and losing significantly in the rural interior. The path to victory for Garza runs through: (1) maximizing turnout in the metro base, (2) running competitively in four key suburban counties, and (3) limiting her losses in rural areas where Whitfield's margins are strongest.

Nadia has been tracking these patterns for months. What she needs from you is to take the analytical framework she has built and extend it into the final 60-day push — which requires a comprehensive analytics plan that ties together the voter file operation, the targeting strategy, the field program, the measurement infrastructure, and the budget.

What Is a Campaign Analytics Plan?

A campaign analytics plan is a living document that specifies: who the campaign is trying to contact, how it is trying to contact them, what it is trying to say to them, how it will measure whether those contacts are working, and what it will do differently if they're not.

The word "living" is important. A plan written 60 days out will be revised at 45 days, at 30 days, at two weeks, and — if the campaign is running well — at one week. New data comes in constantly: early voting reports, updated turnout models, internal poll numbers, field program contact rates, digital ad performance metrics. The analytics plan must be responsive to this data without being destabilized by it. There is a difference between revising a plan in response to clear new evidence and abandoning a plan in response to one bad news day.

Nadia's philosophy of analytics planning: "You build the best model you can on the data you have. Then you update it systematically as you get new data. The model is not the truth — it's the best description of the truth you can get from available evidence. But if you keep updating it honestly, it gets closer."

This is not a different philosophy from anything you've learned in this textbook. It is the same commitment to rigor, honest uncertainty communication, and systematic updating that runs through every chapter. What is different here is the stakes and the clock. The campaign ends on Election Day. There is no going back.

What Nadia Needs from Her New Associate

By the end of your first day, Nadia has outlined her expectations:

Deliverable 1 — Voter universe documentation: A complete description of the campaign's voter universes (persuasion, GOTV, and fundraising), with county-level breakdowns. She wants this in three days.

Deliverable 2 — Contact program projections: For each program (canvassing, phone, mail, digital, text), how many contacts can the campaign realistically make in 60 days, at what cost, targeting which universes?

Deliverable 3 — Measurement framework: What data will we collect, how will we track it, and how will we know if we're on or off track? This includes the weekly metrics Nadia reports to campaign manager Renata Diaz.

Deliverable 4 — Budget allocation recommendation: Given the remaining budget, how should dollars be allocated across programs and geographies?

Deliverable 5 — Ethics and equity review: Before any plan goes to Renata, Nadia wants a documented ethics and equity review. "I've seen campaigns do things with data that I wouldn't be comfortable defending in public. We're not doing those things. I need documentation that we checked."

This capstone is organized around building those five deliverables. By the end, you will have a complete campaign analytics plan — the kind of document that would be handed to a campaign leadership team 60 days before a real Senate race.

Section 2: Voter File Analysis and Universe Building

Understanding the Garza Campaign's Voter File

The voter file is the foundation of everything. Every targeting decision, every contact program, every piece of mail, every door knocked starts from the voter file — the campaign's compiled database of registered voters, enriched with commercial data, field data, and predictive model scores.

The Garza campaign uses VAN (Voter Activation Network) as its primary voter file platform, accessing the state Democratic Party's master voter file. This file contains every registered voter in the state — approximately 3.7 million registrations, of which approximately 2.9 million are estimated to be active (the rest are outdated, deceased, or otherwise inactive). The voter file is matched to commercial data sources that provide demographic estimates (age, income, homeownership, consumer behavior patterns) for voters where public record data is thin.

In addition to the public record data and commercial appends, the Garza campaign has four proprietary data layers:

Layer 1 — Support scores: For each voter, a predicted probability (0-100 scale) that the voter will support Garza on Election Day. These scores are produced by a vendor model trained on prior election data, consumer data, and the campaign's own survey data. Score of 85 means the model predicts an 85 percent probability of Garza support; score of 15 means 85 percent probability of Whitfield support.

Layer 2 — Turnout propensity scores: For each voter, a predicted probability that the voter will cast a ballot in this election. This is modeled from vote history (the single strongest predictor of future turnout), demographic characteristics, and engagement signals. High score = high probability of turning out; low score = low probability.

Layer 3 — Persuadability scores: For each voter, a predicted likelihood that their vote choice is movable — that contact and persuasion from the campaign could shift their support. This is modeled from survey responses, engagement patterns, and characteristics associated with vote switching in prior elections.

Layer 4 — Field-collected data: Survey responses, door-knock results, and phone contact dispositions that the campaign has accumulated over the nine months it has been operating. These are the most reliable data points in the file — actual answers from actual voters — but they cover a small fraction of the total voter file.

The ODA voter file dataset (oda_voters.csv) is a teaching version of this kind of data structure, with columns including voter_id, county, age, gender, race_ethnicity, education, party_reg, vote_history_2018, vote_history_2020, vote_history_2022, urban_rural, support_score, and persuadability_score. You will use this dataset to build and document the campaign's voter universes.

Generating the Persuasion Universe

The persuasion universe is the set of voters the campaign believes it can move — voters who are not firmly committed to either candidate and who are reachable by the campaign's contact programs.

The canonical definition of the persuasion universe uses support score and persuadability score as the primary filters:

Support score filter: Voters with support scores between 40 and 60 are genuinely uncertain — the model gives neither candidate a strong advantage. Scores below 40 are likely Whitfield voters; scores above 60 are likely Garza voters. The persuasion universe targets the middle band.

Persuadability score filter: Within the 40-60 support band, voters with higher persuadability scores are prioritized. Low persuadability even at 50/50 support means the voter is firmly undecided in a way that contact is unlikely to resolve — they may simply be low-information voters who won't be moved by political outreach.

Turnout filter: Persuasion outreach is most cost-effective for voters who are likely to actually vote. Trying to persuade a voter who has a 5 percent turnout probability means your successful persuasion effort has a 5 percent chance of mattering. The persuasion universe typically applies a minimum turnout threshold of 30-40 percent.

Applying these filters to the ODA voter file:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# Load the ODA voter file
voters = pd.read_csv('oda_voters.csv')

print(f"Total voter file: {len(voters):,}")
print(f"By party registration:")
print(voters['party_reg'].value_counts())

# Define persuasion universe
# Support score 40-60: genuinely uncertain
# Persuadability score above 50: movable
# Turnout propensity proxy from vote history
# (In real VAN, turnout propensity score is a separate field;
#  here we construct a proxy from vote history)

def turnout_propensity(row):
    """
    Construct a turnout propensity proxy from three-cycle vote history.
    Returns estimated probability of voting (0-100 scale).
    """
    votes = sum([
        1 if row['vote_history_2018'] == 1 else 0,
        1 if row['vote_history_2020'] == 1 else 0,
        1 if row['vote_history_2022'] == 1 else 0
    ])
    # Weights: 2020 (presidential, high turnout) gets half weight;
    # 2022 (midterm, closer to current cycle) gets double weight
    weighted = (
        (row['vote_history_2018'] * 25) +
        (row['vote_history_2020'] * 15) +
        (row['vote_history_2022'] * 60)
    )
    # Add age adjustment: older voters have slightly higher base turnout
    age_adj = min(10, max(0, (row['age'] - 30) * 0.3))
    return min(100, weighted + age_adj)

voters['turnout_propensity'] = voters.apply(turnout_propensity, axis=1)

# Persuasion universe definition
persuasion = voters[
    (voters['support_score'] >= 40) &
    (voters['support_score'] <= 60) &
    (voters['persuadability_score'] >= 50) &
    (voters['turnout_propensity'] >= 30)
].copy()

print(f"\nPersuasion universe: {len(persuasion):,}")
print(f"As % of total voter file: {len(persuasion)/len(voters)*100:.1f}%")

# Priority tiers within persuasion universe
persuasion['priority_tier'] = 'Tier 3 (Low)'
persuasion.loc[
    (persuasion['turnout_propensity'] >= 60) &
    (persuasion['persuadability_score'] >= 65),
    'priority_tier'
] = 'Tier 1 (High)'
persuasion.loc[
    persuasion['priority_tier'].str.contains('Low') &
    (persuasion['turnout_propensity'] >= 45),
    'priority_tier'
] = 'Tier 2 (Medium)'

print("\nPersuasion universe by priority tier:")
print(persuasion['priority_tier'].value_counts())

# County-level breakdown
county_persuasion = persuasion.groupby('county').agg(
    universe_size=('voter_id', 'count'),
    avg_support=('support_score', 'mean'),
    avg_persuadability=('persuadability_score', 'mean'),
    avg_turnout_propensity=('turnout_propensity', 'mean'),
    tier1_count=('priority_tier', lambda x: (x == 'Tier 1 (High)').sum())
).reset_index()

county_persuasion['tier1_pct'] = (
    county_persuasion['tier1_count'] / county_persuasion['universe_size'] * 100
)

print("\nTop 10 counties by persuasion universe size:")
print(county_persuasion.nlargest(10, 'universe_size')[
    ['county', 'universe_size', 'avg_support', 'tier1_count', 'tier1_pct']
].to_string(index=False))

The output of this analysis defines the campaign's persuasion universe: the voters Garza's campaign will target with persuasion messaging across its field, digital, and mail programs. A typical mid-scale Senate campaign might have a persuasion universe of 150,000 to 300,000 voters — large enough to require systematic contact programs but small enough to be achievable within budget.

Generating the GOTV Universe

The GOTV (Get Out the Vote) universe is distinct from the persuasion universe in a crucial way: it targets voters who are already likely to support Garza but whose turnout is uncertain. You are not trying to persuade them — you are trying to ensure they vote.

GOTV universe definition:

Support score filter: Support score above 65 — the voter is likely a Garza supporter.

Turnout propensity filter: Turnout propensity between 30 and 75 — high enough that GOTV contact might push them to vote, low enough that they are genuinely at risk of not voting without contact. Very low propensity voters (below 30) are expensive to turn out and may not be reachable. Very high propensity voters (above 75) will likely vote with or without contact — GOTV resources are wasted on them.

Party registration modifier: Registered Democrats in the GOTV universe receive the highest priority. Independents with high Garza support scores receive secondary priority.

# GOTV universe definition
gotv = voters[
    (voters['support_score'] >= 65) &
    (voters['turnout_propensity'] >= 30) &
    (voters['turnout_propensity'] <= 75)
].copy()

print(f"GOTV universe: {len(gotv):,}")
print(f"As % of total voter file: {len(gotv)/len(voters)*100:.1f}%")

# Priority weighting for GOTV
# Most valuable: high support, medium turnout (50-65 propensity)
gotv['gotv_priority'] = 'Standard'
gotv.loc[
    (gotv['support_score'] >= 75) &
    (gotv['turnout_propensity'].between(45, 70)),
    'gotv_priority'
] = 'High'
gotv.loc[
    (gotv['party_reg'] == 'Democrat') &
    (gotv['gotv_priority'] != 'High'),
    'gotv_priority'
] = 'Party Base'

print("\nGOTV universe by priority:")
print(gotv['gotv_priority'].value_counts())

# Urban/rural breakdown — critical for GOTV program design
print("\nGOTV universe by urban/rural classification:")
print(gotv['urban_rural'].value_counts())

# Demographic breakdown for message customization
print("\nGOTV universe by race/ethnicity:")
print(gotv.groupby('race_ethnicity')['voter_id'].count().sort_values(ascending=False))

One of the most important analytical findings from the GOTV analysis concerns the racial and ethnic composition of the GOTV universe. In a state where 32 percent of the electorate is Latino and 18 percent is Black, the GOTV universe skews heavily toward these communities — because Latino and Black voters in this state show higher support for Garza but more variable turnout patterns. This has profound implications for program design, which we'll address in Section 4.

Generating the Fundraising Universe

The fundraising universe targets voters and supporters who are likely to make financial contributions to the campaign. Unlike the persuasion and GOTV universes, the fundraising universe is not primarily drawn from the voter file — it draws heavily from the campaign's existing donor database (from the oda_donations.csv dataset) and modeled from donor characteristics.

The fundraising universe includes three sub-segments:

Prior donors: Anyone who has contributed to Garza in any cycle. These are the campaign's best prospects for additional giving.

Lapsed donors: People who gave to Democratic candidates in prior cycles but have not yet contributed to Garza. They are warm prospects — the campaign just hasn't asked them yet (or hasn't asked them effectively).

New prospect donors: Voter file records that match the demographic profile of the campaign's donor base — high education, high income, professional employment, consistent voting history. These are cold prospects but worth systematic outreach.

The fundraising universe analysis uses the oda_donations.csv dataset to characterize the donor base's demographic and geographic patterns:

import pandas as pd

donations = pd.read_csv('oda_donations.csv')

# Filter to Garza contributions
garza_donors = donations[
    donations['recipient'].str.contains('Garza', na=False)
].copy()

print(f"Total Garza contributions: {len(garza_donors):,}")
print(f"Unique donors: {garza_donors['donor_name'].nunique():,}")
print(f"\nContribution breakdown by type:")
print(garza_donors['donation_type'].value_counts())

# Average and median contribution size
print(f"\nContribution amounts:")
print(garza_donors['amount'].describe())

# Geographic distribution
print("\nTop donor states:")
print(garza_donors['donor_state'].value_counts().head(10))

# In-state vs. out-of-state breakdown
in_state_pct = (garza_donors['donor_state'] == garza_donors['donor_state'].mode()[0]).mean() * 100
print(f"\nEstimated in-state donor share: {in_state_pct:.1f}%")

# Occupation breakdown for prospect modeling
print("\nTop donor occupations:")
print(garza_donors['occupation'].value_counts().head(15))

The fundraising universe analysis informs the analytics plan's recommendation that fundraising outreach be concentrated in the campaign's final three weeks rather than evenly distributed — research consistently shows that urgency cues (deadline-driven asks) are more effective in the campaign's final stretch, and that donor fatigue sets in when campaigns ask too frequently throughout the cycle.

Universe Overlap and Budget Allocation

A voter can appear in multiple universes. A high-support, medium-turnout voter who has previously donated might appear in both the GOTV universe and the fundraising prospect list. A voter at exactly 50/50 support with high persuadability and moderate turnout might appear in both the persuasion universe and the GOTV universe (the support score boundary is not perfectly precise).

Universe overlap matters for budget allocation because contacting a voter with a GOTV message when they need persuasion is wasteful — you're deploying the wrong message for the voter's actual situation. Contacting a persuasion target with a GOTV-style "make sure you vote" message before you've won them over is premature at best and counterproductive at worst (you're reminding someone to vote who might vote against you).

Nadia's universe management protocol specifies:

1. In the first 30 days (60-30 days out), persuasion universe contacts take priority. The goal is to move undecided voters before GOTV intensity increases.
2. In the middle 15 days (30-15 days out), persuasion and GOTV run simultaneously at equal intensity.
3. In the final 15 days, GOTV dominates. Persuasion contacts continue only for Tier 1 persuasion targets (highest priority).
4. During the early voting period (which begins 14 days before Election Day in this state), all contacts receive a ballot-request or early voting location message regardless of their universe designation.

The budget allocation across universes should follow this timeline logic: more persuasion budget in the first half, more GOTV budget in the second half, with digital program flexibility maintained throughout.

County-Level Universe Breakdowns

The analytics plan includes a complete county-level breakdown of all three universes. This breakdown drives the geographic allocation of field resources (where to deploy canvassers and phone banks), the geographic targeting of mail and digital programs, and the county-level performance benchmarks used in the measurement framework.

The county-level analysis reveals several strategic priorities:

County A (Major Metro, Garza's base): Very large GOTV universe, moderate persuasion universe. The path here is maximizing Garza base turnout. A four-point increase in Garza margin in this county alone is worth approximately 18,000 net votes statewide.

County B (Major Metro Suburban Ring, swing county): Large persuasion universe, significant GOTV universe. This is the campaign's highest strategic priority. College-educated women here are the most persuadable demographic in the state. Losing County B by more than five points likely means losing the race.

County C (Smaller Metro, high Latino share): Large GOTV universe concentrated in Latino communities, moderate persuasion universe. Garza won County C in the primary by 28 points; the general requires a similar margin but with significantly lower expected base turnout. GOTV program quality in County C is one of the top three strategic levers in the race.

County D through F (Rural interior, Whitfield country): Small Garza universes, significant Republican advantage. The analytics plan does not recommend significant investment in these counties — Garza cannot win them and is not trying to — but the plan flags any Garza-supporting voter in these counties with an intensive GOTV designation because turning out an isolated Garza supporter in a rural county requires extra effort that mail and digital alone cannot provide.

Section 3: Targeting Strategy and Message Matrix

Demographic Targeting Priorities

Garza's targeting strategy is organized around four demographic priority segments, each requiring a different message emphasis and contact approach.

Segment 1 — Latino Voters (32% of electorate)

Garza's strongest demographic, but with significant internal heterogeneity. First-generation immigrant voters, U.S.-born second-generation voters, and third-plus-generation Latino voters have meaningfully different issue priorities and information environments. First-generation voters prioritize immigration policy and economic security. Second-generation voters are more engaged with civil rights, healthcare access, and education. Third-plus-generation Latino voters — who are most likely to be registered and to have consistent vote histories — are more heterogeneous ideologically and require individualized message testing rather than demographic assumption.

Latino voter turnout in midterm cycles is historically lower than presidential year cycles, making GOTV investment in this segment particularly important. Garza's support among this segment in Meridian's most recent internal poll is 68 percent — strong but not overwhelming, and consistent with historical Democratic performance in the state.

Segment 2 — Black Voters (18% of electorate)

Garza's most reliable base, with support running approximately 85-90 percent in internal polling. The strategic priority with this segment is entirely about turnout — persuasion investment is essentially not needed, while GOTV investment is among the campaign's highest returns on investment. Research from prior election cycles in the state shows a strong correlation between Black voter turnout in specific precincts and the effectiveness of community-based GOTV programs (as opposed to generic mail or phone banking). The analytics plan prioritizes community-based, trusted-messenger GOTV outreach in majority-Black precincts.

Segment 3 — College-Educated White Women in Suburban Counties

This segment is the heart of the campaign's persuasion universe. Internal polling shows Garza running approximately 55-45 among this segment — competitive but not commanding. The segment has a history of ticket-splitting and late decision-making, making it genuinely persuadable. Message testing has identified healthcare cost (particularly prescription drug costs and insurance coverage) and reproductive rights as the highest-performing messages with this segment. Environmental concerns are secondary but significant.

This segment is also high-information: they consume news regularly, are skeptical of advertising claims, and respond to evidence-based messaging more than emotional appeals. The analytics plan recommends direct mail (which this segment reads) and digital video (which they watch) over phone contact, which this segment consistently rates as intrusive.

Segment 4 — Young Voters (Ages 18-29)

A high-variance, high-stakes segment. Young voters (18-29) in the ODA voter file show disproportionately low vote history (a majority of voters in this age range have zero or one cycle of vote history) but disproportionately high support for Garza among those who do engage with the campaign. The challenge is the turnout gap: if young voters voted at the same rate as voters over 45, Garza's margin in the state would increase by approximately 3.5 points based on current support distributions.

The analytics plan recommends a text messaging and peer-to-peer outreach strategy for young voters, prioritizing campus-based organizing in the state's three major university communities, and digital advertising on platforms with high young adult reach. Young voter outreach requires earlier timing than other GOTV programs — research shows that young voters who receive early information about how to vote (registration deadlines, polling locations, early voting options) are more likely to follow through than those who receive reminders close to Election Day.

Geographic Targeting: Where to Deploy Resources

Geographic targeting begins with the county-level universe analysis from Section 2 and adds a cost-effectiveness overlay: not all contacts are equally expensive to make. Urban contacts (dense canvassing territory, shorter drive times, larger phone bank pools) are cheaper per unit than rural contacts. High-propensity voters in dense urban precincts may be the cheapest contacts in the entire universe.

The geographic resource allocation model uses a simple expected-value framework:

Expected value of county investment = (persuasion universe size × expected persuasion rate × persuasion lift) + (GOTV universe size × expected contact rate × turnout lift)

Where: - Expected persuasion rate: the fraction of the persuasion universe the campaign can realistically contact given budget - Persuasion lift: the estimated increase in support probability from a successful persuasion contact (based on field experiment evidence, typically 1-3 percentage points) - Expected contact rate: the fraction of the GOTV universe the campaign can contact - Turnout lift: the estimated increase in turnout probability from a successful GOTV contact (based on field experiment evidence, typically 2-8 percentage points depending on contact type)

This framework generates a ranked list of counties by expected marginal impact per dollar spent. The analytics plan uses this ranking to allocate field staff, phone banking capacity, and mail and digital budget across the state.

One important complication: the expected-value framework is a guide, not a dictator. There are strategic reasons to invest in counties even when the expected-value calculation suggests otherwise. County B (the suburban swing county) may be the highest expected-value target by far — and it is — but even if it weren't, its symbolic importance and media market salience might justify above-model investment.

Message Matrix: Which Messages Go to Which Voter Segments

The message matrix is one of the analytics plan's most important deliverables. It specifies, for each voter segment, which campaign messages are deployed, in which order, and through which channels.

Garza's campaign has three primary message tracks in the final 60 days:

Track 1 — Healthcare: Focuses on Garza's record on healthcare access, prescription drug costs, and protecting coverage for pre-existing conditions. This is the strongest performing message with suburban college-educated voters and with voters who have personal or family healthcare cost concerns.

Track 2 — Economic Security: Focuses on Garza's record on fair wages, housing affordability, and economic opportunity. This is the strongest performing message with Latino working-class voters, Black working-class voters, and younger voters anxious about economic prospects.

Track 3 — Accountability and Integrity: Focuses on Garza's prosecutorial record, her commitment to public accountability, and contrast with Whitfield on corruption concerns. This message performs well with voters who are skeptical of both parties and are looking for a candidate who will hold power accountable.

The message matrix assigns primary and secondary tracks to each segment:

Message sequencing also matters. In the persuasion program, Nadia's preferred sequence is: introduce (who is Maria Garza?) → contrast (how does she differ from Tom Whitfield on the issues you care about?) → close (here's what's at stake and why your vote matters). A voter who receives a strong contrast message before they have any positive context for Garza is less persuadable than one who first gets the positive introduction.

The analytics plan specifies that persuasion mail should follow a four-piece sequence over 45 days (pieces at 45 days out, 35 days out, 25 days out, and 15 days out), with digital advertising running concurrently. GOTV contacts begin in earnest at 20 days out.

Digital vs. Field Deployment Decisions

The channel allocation question — how much of the persuasion and GOTV contact program should run through digital advertising vs. field (canvassing and phone) — is one of the most contested strategic questions in modern campaign analytics.

The research on comparative effectiveness is unambiguous about a few things:

Door-to-door canvassing is the single most effective voter contact method for increasing turnout. Studies consistently find turnout lifts of 4-8 percentage points per successful canvass contact — substantially higher than phone, mail, or digital. But canvassing is also the most expensive method per contact, requires volunteer infrastructure, and is geographically limited to areas where volunteers can be deployed.

High-quality phone contact (live calls, not automated robocalls) is moderately effective for GOTV (turnout lifts of 2-4 points per successful contact) and for persuasion (persuasion lifts of 1-2 points per successful contact). It is more cost-efficient than canvassing but less effective per contact.

Mail is the most reliable in terms of delivery (a piece of mail that enters the postal system almost always reaches its target) but the least reliable in terms of impact (read rates are uncertain, and mail cannot be personalized in real time). Mail is most valuable for high-information segments who read and engage with direct mail, and as a baseline presence in segments where field and digital access is limited.

Digital advertising is the most cost-efficient method for reach — digital ads can reach large audiences at pennies per impression — but has the most contested evidence on effectiveness. The most credible studies of digital advertising's persuasion and GOTV effectiveness find modest effects (persuasion lifts of 0.3-0.8 points, turnout lifts of 0.5-2 points per effective exposure). Digital's strength is reach and frequency, not per-contact impact.

Nadia's allocation philosophy: "Use field for your highest-priority targets, where the high per-contact effectiveness justifies the high per-contact cost. Use digital for broad-reach programs where you're trying to maintain presence across a large audience you can't reach with field. Use mail for high-information segments and for geographic areas where field coverage is thin. Use phone for the middle tier — people you want to contact but can't canvas."

The Microtargeting Ethics Discussion

Nadia has an internal debate that she shares with you explicitly. The voter file's predictive model scores and commercial data appends enable targeting at a level of individual specificity that raises genuine ethical questions, even in a campaign context.

Specifically: the campaign can target advertising to voters based on commercially purchased data about their medical conditions (inferred from prescription purchase records and consumer behavior), their financial situation (inferred from credit and consumer data), their religious practice (inferred from consumer data), and their personal relationships (inferred from social media behavioral data). This information enables hyper-personalized messaging — a voter inferred to have a family member with diabetes gets a very specific healthcare message; a voter inferred to be financially stressed gets a specific economic message.

Nadia's concern is not that this targeting is illegal — it is legal — but that it represents a kind of asymmetric surveillance that voters have not consented to. "If I told a voter on the doorstep: 'I know your family member has diabetes because I bought consumer data that predicted it, and I've targeted this message specifically to that fact' — how would they feel about that? Would they feel served, or surveilled?"

The ethics discussion is documented in the analytics plan as follows:

What the campaign does: Target messaging using demographic characteristics (age, gender, geography), vote history, and standard consumer data segments (education level, income bracket, homeownership). Test messages for effectiveness through randomized experiments. Personalize GOTV messages with individual voter's polling location and early voting information.

What the campaign does not do: Target messages based on inferred health conditions, financial distress indicators, or relationship status derived from consumer data. Use lookalike audience modeling that matches voter data to social media platform behavioral profiles. Purchase third-party data on religious practice or political views inferred from private communications.

The campaign's analytics plan is explicit that these prohibitions apply even in cases where the prohibited techniques might be effective. Effectiveness is not the only metric that matters.

Section 4: Field Program Design

Canvassing Plan: Universe, Script, Volunteer Targets, Timeline

The Garza campaign's canvassing program is the backbone of its GOTV and persuasion operations. Canvassing requires: a well-defined turf (geographic territory assigned to each canvasser), a pre-loaded walk list (the specific voters to contact in that turf), a script (what the canvasser says and what data they collect), and a training program (how to prepare volunteers to execute this effectively).

Turf management: The analytics plan uses a geographic information system to divide the campaign's target counties into canvassing turfs — clusters of 25-40 doors that a single canvasser can work in a three-hour shift. Turf boundaries account for walkability (houses close enough together to minimize drive time) and universe density (enough target voters per block to make the walk efficient). Turfs in urban and suburban areas are walk-efficient; rural turfs require driving between stops and are allocated to volunteers with vehicles.

Walk list generation: The walk list is drawn from the voter file universes established in Section 2. GOTV turfs prioritize high-support, medium-turnout voters. Persuasion turfs prioritize high-persuadability, medium-support voters. In mixed turfs (which are the norm in suburban areas), canvassers receive universe flags for each voter on their list so they know which message to deliver.

Script design: The analytics plan recommends two primary scripts, field-tested in the campaign's primary:

GOTV script (for base voters): Personal, warm, information-focused. "Hi, I'm [name], a volunteer for Maria Garza's campaign. I'm here to help make sure every voter in the neighborhood knows about the election on [date] and has everything they need to vote. Will you be voting this November?" [Branch to: early voting information / mail ballot information / polling place information based on voter preference.] "We're counting on voters like you — thank you for your time."

Persuasion script (for undecided voters): Discovery-first, not assertion-first. "Hi, I'm [name], a volunteer for Maria Garza's campaign. I know you're probably hearing a lot from both sides. I'm not here to tell you who to vote for — I want to know what's most important to you this election." [Listen and identify top issue.] [Deliver relevant message from message matrix.] "Is there anything about Maria Garza's position on [issue] you'd like more information about?" [Offer literature and follow-up contact.]

Volunteer targets: The canvassing plan requires approximately 2,400 trained volunteer-shifts (a shift being three hours of active canvassing) to complete the field program. Recruitment targets are set county by county based on the local volunteer infrastructure and universe size. The analytics plan tracks volunteer recruitment weekly against these targets and includes contingency plans if recruitment falls short: converting portions of the canvassing program to phone banking (fewer volunteers needed), prioritizing highest-expected-value turfs, and in extreme shortage scenarios, supplementing with paid canvassers in the highest-priority areas.

Timeline: The canvassing program runs from day 60 to day 1 (Election Day itself). Persuasion canvassing is heaviest in weeks 8-4 (days 60-30). GOTV canvassing is heaviest in the final two weeks (days 14-1). A special program targets early voters beginning on the first day of early voting.

Phone Program: Volunteer vs. Paid, CATI vs. Relational

The phone program supplements canvassing in geographic areas where field coverage is thin, for voter segments where phone contact has shown effectiveness, and for follow-up contacts with voters who have been canvassed but require a second touch.

The analytics plan recommends a hybrid phone program:

Volunteer phone banking: Structured around evening and weekend calling sessions at campaign offices, using VAN's phone banking interface. Best for: initial GOTV contacts with the campaign's base, volunteer cultivation, supporter ID calls for fundraising. Volunteer phone banking suffers from inconsistent call quality and early session fatigue (volunteers make great calls for the first hour, mediocre calls after), but it is essentially free beyond staff coordination time.

Relational organizing: Each campaign volunteer is encouraged to contact 10-15 people in their personal network using the campaign's relational organizing tool (a digital platform that allows volunteers to look up their contacts in the voter file and record their conversations). Relational organizing produces dramatically higher contact rates and persuasion effectiveness than stranger-to-stranger phone banking — a call from someone you know is returned at two to three times the rate of a call from a stranger. The analytics plan sets a relational organizing target of 8,000 personal network contacts across the campaign's volunteer base.

Paid CATI (Computer-Assisted Telephone Interviewing): For survey calls — measuring support scores, testing messages, and tracking campaign performance metrics — the campaign contracts with Meridian Research Group for formal CATI surveys. These are not persuasion or GOTV calls; they are data collection instruments. The polling plan (Section 5) describes this in detail.

Mail Program: Targeting, Timing, Message Sequencing

The mail program runs approximately 1.1 million pieces of mail across the campaign's final 45 days, covering both the persuasion universe and the GOTV universe.

Persuasion mail: Four-piece sequence to Tier 1 and Tier 2 persuasion universe voters (estimated 175,000 unique voters receiving the sequence). Pieces are: - Piece 1 (45 days out): Introduction / biography piece — who is Maria Garza and what is her record? - Piece 2 (35 days out): Healthcare contrast piece — how do the candidates differ on healthcare? - Piece 3 (25 days out): Economic security contrast piece — jobs, housing, wages - Piece 4 (15 days out): Close / urgency piece — what's at stake, how to vote

GOTV mail: Two-piece sequence to GOTV priority voters. Piece 1 (20 days out) provides early voting information and personal polling location details. Piece 2 (7 days out) is a vote reminder with ballot tracking information (if the state offers it) and an urgency message.

Targeted specialty mail: Spanish-language versions of all pieces for Spanish-surname voters in the persuasion and GOTV universes. A separate healthcare-focused piece targeting voters with health-related consumer data flags (subject to the ethics constraints in Section 3). A youth-focused piece targeting 18-29 voters with campus or urban addresses.

Mail testing: The analytics plan recommends a split test on Pieces 2 and 3 — two versions of each piece with different design approaches — allocated randomly within matched targeting universes. Response rates cannot be directly measured for mail, but the campaign tracks early vote rates and IDs among mail recipients to test whether mail sequence variations correlate with different outcomes.

Text/SMS: Opt-in List Building, Compliance

Text messaging is one of the most effective voter contact methods for young voters and for voters who have expressed interest in the campaign. It is also the most legally regulated contact method: federal law (the Telephone Consumer Protection Act) requires prior express written consent before sending campaign text messages to mobile phones using automated messaging systems.

The Garza campaign's text program operates entirely from its opt-in list — voters who have affirmatively signed up to receive campaign texts through the website, at events, through volunteer recruitment, or through digital advertising campaigns that include a text opt-in call to action.

The analytics plan projects that the opt-in list will reach approximately 42,000 numbers by the final 30 days of the campaign. Texts are targeted within this list by: voter's universe designation (GOTV vs. persuasion messages), prior engagement (has the voter responded to previous texts?), and geography (event announcements are geographically targeted).

Compliance requirements: All texts include an opt-out instruction ("Reply STOP to unsubscribe"). All texts are logged with timestamp and content. Opt-out requests are processed within 24 hours. The campaign does not use peer-to-peer texting tools to circumvent TCPA restrictions — all automated messages go to opt-in contacts only.

Integrating Digital and Field

Digital advertising and field programs must be coordinated to avoid redundancy and maximize the cumulative effect of multi-channel contact.

The analytics plan uses a "coordinated contact" model: voters in the Tier 1 persuasion universe receive both digital advertising and at least one field contact (canvassing or phone) within a 10-day window. Research suggests that multi-channel contact (seeing a campaign's message in multiple formats across multiple channels) produces higher persuasion lifts than single-channel contact at equivalent cost. The challenge is coordination: knowing which digital targets have also been contacted in the field, and vice versa, requires that the digital ad targeting file (derived from the voter file) and the field contact database (from VAN) be linked and updated regularly.

Nadia's integration protocol: digital targeting lists are refreshed weekly from the updated voter file universe. Voters who have been successfully canvassed (recorded as a "persuasion contact" in VAN) are moved from the persuasion digital pool to the GOTV digital pool in the weekly refresh. This prevents the campaign from running persuasion ads to voters who have already committed to Garza, and ensures GOTV digital reinforces the field GOTV program rather than running parallel to it.

Section 5: Polling and Research Plan

Internal Polling Cadence and Budget

Internal polling serves a different purpose than public polling. Public polls (conducted by Meridian and other firms and released to media) measure the race's standing for external audiences: media, donors, voters, and the broader political ecosystem. Internal polls are designed for the campaign's own decision-making.

The Garza campaign's internal polling budget for the final 60 days is allocated as follows:

Benchmark survey (60 days out): Full-race survey, N=800, with complete issue priority, message testing, and candidate favorability battery. This is the baseline against which all subsequent movement is measured. Commissioned from Meridian Research Group.

Tracking surveys (every three weeks): Shorter surveys (N=500-600) measuring head-to-head horse race numbers, candidate favorability, and a rotating subset of issue questions. Three tracking surveys are planned at 45, 30, and 15 days out.

Message tests (embedded in tracking):The 45-day and 30-day tracking surveys each include a split-sample message test — half the respondents are read Message A, half Message B, on a specific contrast question. This allows the campaign to test which framing is more effective before committing advertising budget.

Targeted county polling: One dedicated county-level survey (N=400) in County B (the suburban swing county) at 25 days out, providing county-specific numbers to guide field resource allocation.

Total polling budget: approximately $185,000 — a significant investment that the analytics plan justifies to campaign leadership in Section 7.

What to Poll For: Horse Race vs. Message Testing vs. Opposition Research

Campaign polling is never just about the horse race number (Garza X%, Whitfield Y%). A campaign that only uses internal polling to track the head-to-head number is wasting most of the value of the survey.

The Garza campaign's internal surveys track:

Candidate favorability and unfavorability: Both positive and negative ratings for Garza and Whitfield. Unfavorability trends are often more predictive than favorability — voters rarely switch from strongly favorable to strongly unfavorable, but voters who are already lukewarm can be pushed to unfavorable.

Issue salience: "What is the most important issue in deciding your vote?" Tracking this over time shows whether the campaign's issue emphasis is aligned with voter priorities or out of phase.

Message testing: Which of two framings of a contrast between the candidates is more effective at moving the head-to-head number? Message testing requires a split-sample design (two randomly assigned groups, each seeing a different message) and careful question sequencing.

Candidate attribute ratings: Beyond overall favorability, how do voters rate each candidate on specific attributes — trustworthy, understands people like me, strong leadership, handles healthcare, handles the economy? Attribute tracking shows which impressions of the candidates are most movable and where the campaign is succeeding or falling short.

Likely voter screening: The composition of the likely voter screen — which voters are likely to participate — changes as the campaign progresses. Tracking how the likely voter screen shifts allows the campaign to assess whether its GOTV program is expanding the electorate in the targeted directions.

Working with Meridian Research Group

The Garza campaign's relationship with Meridian Research Group is professional but not uncomplicated. Meridian is a nonpartisan firm — they also poll for Whitfield's campaign — which means the data they provide to Garza is the same quality and methodology as what they provide to Whitfield. This is actually reassuring: Meridian's methodological reputation means their internal surveys will not be designed to please the client.

Dr. Vivian Park at Meridian has established clear protocols for client campaigns: she presents findings as they are, not as clients wish they were; she reports uncertainty honestly (a 46-44 lead with a 4-point margin of error is not a "two-point lead" — it is a statistical tie); and she will decline to conduct a survey if the client specifies design choices she believes would compromise the data's integrity.

Carlos Mendez, Meridian's junior analyst, serves as the primary day-to-day contact for the Garza campaign account. He briefs Nadia after each survey completion, walks through the crosstabs with her, and flags findings that he thinks deserve more attention than the top-line numbers suggest.

The analytics plan specifies that Nadia will attend the debrief for every internal survey and will summarize the findings for campaign manager Renata Diaz in a two-page written brief within 48 hours of receiving the data. The brief always includes: the headline finding, the most important subgroup finding, any movement since the last survey, and Nadia's interpretation of what the data means for the campaign's strategy.

Interpreting Internal Polls vs. Public Polls

Internal polls and public polls often tell different stories. Managing the gap between them is one of the trickier parts of a campaign analytics director's job.

Internal polls can show a lead larger than public polls for several legitimate reasons: the campaign's likely voter screen may be more optimistic about its base turnout than public polls assume; the campaign's survey may have been in the field during a favorable news cycle; or the campaign may be polling on a slightly different question framing.

Internal polls can also show a lead larger than public polls for illegitimate reasons: the methodology may be biased toward finding favorable results (leading questions, bad sampling); the campaign may selectively share favorable internal results; or staff may be gaming the survey design to show the boss what the boss wants to see.

Nadia's protocol for managing this tension: she tracks Garza's position in public polling aggregates (using a standard weighted average of available public polls) alongside the internal polling results, and documents the gap. If the internal poll shows Garza up by 5 when the public aggregate shows a 2-point lead, Nadia treats the 2-point aggregate number as the better estimate of reality and uses the internal poll to understand why the gap exists.

"I've been in campaigns that convinced themselves they were winning because they trusted their own internals over the aggregate," Nadia tells you. "They were wrong every time. Your internal polls are a diagnostic tool, not a comfort tool."

Section 6: Measurement and Evaluation Framework

KPIs for Each Program Component

The analytics plan's measurement framework begins with Key Performance Indicators (KPIs) for each contact program. KPIs are not goals — they are metrics. The distinction matters: a goal is a desired outcome (win the race); a KPI is a measurable indicator that helps assess whether you're on track to reach the goal.

Canvassing KPIs: - Contact rate (doors knocked / doors answered, target: 25-35% for GOTV, 20-30% for persuasion in urban/suburban areas) - Supporter ID rate (of contacts made, what fraction identified as Garza supporters, target: 60-70% in GOTV universe) - Turf completion rate (% of assigned turfs fully worked, target: 85%+) - Volunteer retention rate (% of trained volunteers who complete at least 3 shifts, target: 60%+)

Phone banking KPIs: - Contact rate (calls placed / live contacts, target: 10-15% for cold calls) - Conversation completion rate (% of contacts who complete the full script, target: 45-55%) - Supporter ID distribution (tracking support tiers for universe refinement)

Mail KPIs: Mail lacks direct response metrics, so KPIs are indirect: - Delivery rate (% of pieces returned undeliverable, target: under 8%) - Early vote rate among mail recipients (tracked through early vote file matches, compared against non-mail universe)

Digital advertising KPIs: - Cost per click (for digital ads with a direct response component, target by placement type) - Video completion rate (for video ads, target: 40-55% full video completion) - Cost per acquisition (for volunteer recruitment digital ads, cost per new volunteer signup) - Audience match rate (% of voter file universe successfully matched to digital ad platforms, target: 60-75%)

Text KPIs: - Delivery rate (% of texts delivered, target: 95%+) - Response rate (for two-way text conversations, target: 15-25%) - Opt-out rate (target: under 3% per send)

Overall program KPIs: - Total contacts made (cumulative across all channels, tracked against universe goals) - Universe penetration rate (% of priority universes with at least one contact, targets: 85% of Tier 1 persuasion, 70% of Tier 1 GOTV) - Contact-to-commitment rate (% of contacts who commit to support or commit to vote, tracked across channel for relative effectiveness comparison)

The Dashboard Applied to This Campaign

The voter contact dashboard developed in Chapter 33 provides the technical infrastructure for the Garza campaign's real-time monitoring. Nadia's dashboard displays:

Daily view: New contacts logged by program (canvassing, phone, text, digital response), running total against universe targets, geographic distribution of contacts by county, and daily early vote/mail ballot return rates (once early voting begins).

Weekly view: Universe penetration rates by county and priority tier, program-specific KPIs with trend lines, volunteer recruitment and retention metrics, polling summary (when a survey has been completed), and budget pacing (spending rate vs. plan).

Strategic view (updated weekly by Nadia for campaign leadership): Projected final contact totals by universe, estimated vote margin based on polling plus contact program performance, county-level performance flags (counties where contact targets are at risk), and the week's key decision items.

The dashboard is not a vanity metric display — it is a decision-support tool. Its value comes not from what it shows but from what it enables: catching problems early enough to fix them. A canvassing contact rate that falls below target in week 4 means either the turf lists are wrong (too few target voters per block), the volunteer training is inadequate, or there's an external factor suppressing response rates. Catching that problem at week 4 leaves three weeks to diagnose and correct it; catching it at week 8 leaves no time.

A/B Testing Opportunities

The analytics plan identifies four A/B testing opportunities within the contact program:

Test 1 — Mail piece design: Two design versions of Persuasion Mail Piece 2 (the healthcare contrast piece), randomized at the voter level within matched propensity strata. Outcome measurement: support ID rates in subsequent field contacts. Expected sample size: 40,000 per arm.

Test 2 — Canvassing script variation: Two script versions for persuasion canvassing (a "discovery-first" script and a "message-first" script), randomized at the turf level. Outcome measurement: supporter ID rate at the door and, for voters canvassed, early vote rates. Expected sample size: 200 turfs per arm.

Test 3 — GOTV text message timing: Early GOTV text (20 days out) vs. late GOTV text (10 days out) for matched voter pairs in the GOTV universe. Outcome measurement: early vote return rate. Expected sample size: 5,000 per arm.

Test 4 — Digital ad message frame: Positive (Garza's record and vision) vs. contrast (Garza vs. Whitfield on healthcare) in digital video advertising to the persuasion universe. Outcome measurement: click-through rate and any available downstream survey evidence. Expected sample size: 50,000 impressions per arm.

All four tests are registered in the campaign's analytics documentation before launch, with pre-specified hypotheses and analysis plans. This registration prevents post-hoc fishing — running many tests and only reporting the ones that "work" — and creates accountability for the analytics team.

Post-Election Evaluation

The analytics plan is written before the election. It will be evaluated after the election in a formal postmortem — standard practice for campaigns that take data seriously.

The postmortem will assess:

Prediction accuracy: How well did the voter universes reflect the actual distribution of Garza voters? The actual election returns can be matched to the voter file to assess the support score model's calibration — were voters scored at 65 actually Garza supporters 65 percent of the time?

Contact program effectiveness: Do early vote and final vote rates differ significantly between contacted and non-contacted voters in the same universe tier? This comparison (using the campaign's contact database matched against the final official voter file) is the most direct evidence available on whether the contact program moved votes.

Turnout model accuracy: Did voters with high turnout propensity scores actually turn out at higher rates than predicted? Did the GOTV program's effect show up as higher-than-predicted turnout among contacted voters?

Geographic accuracy: Did the county-level results track the model predictions? Which counties beat the model, and which underperformed? Geographic model error points to systematic issues with the underlying data or modeling assumptions.

Nadia commits to writing a postmortem regardless of the election outcome. "Win or lose, I want to know what the data actually showed. If we won, I want to know which parts of the plan made the difference. If we lose, I want to know what we missed. The only way to get better is to be honest about what the evidence says."

Nadia's Weekly Reporting Cadence

Every Monday morning, Nadia delivers a two-page written brief to campaign manager Renata Diaz. The brief format is fixed:

Section 1 — Numbers from the week (half page): Key metrics from the previous week's contact program. No interpretation — just numbers against targets.

Section 2 — What the numbers mean (half page): Nadia's interpretation. What's working, what's not, what's changed since last week, what the trend line suggests.

Section 3 — Decisions needed (half page): Specific decisions that require Renata's input. These are framed as options with expected outcomes: "We can shift $40,000 from the phone program to canvassing in County B, which the data suggests would increase our contact rate there by approximately 15 points. The tradeoff is reducing phone coverage in Counties D and E, where we have a smaller universe. My recommendation is to make this shift. Your decision."

Section 4 — What to watch (quarter page): Early indicators of developing trends. Things that are not yet at the decision threshold but need monitoring.

The brief always ends with a sentence stating the current best estimate of the race outcome and what would need to change in either direction. Nadia's commitment to honest uncertainty communication — learned partly from her work with Meridian — means she never rounds a 2-point lead up to "we're winning comfortably" and never rounds a 2-point trail down to "we're in serious trouble." She calls it as she sees it, with appropriate confidence intervals.

Section 7: Budget and Resource Allocation

A Realistic Campaign Analytics Budget

The Garza campaign's analytics-related budget for the final 60 days is $1.12 million, allocated as follows. These figures represent realistic costs for a competitive U.S. Senate race in a mid-sized state.

Voter Contact Programs:

Research and Polling:

Technology and Data:

Staff:

Total analytics budget: $1,120,000

This budget allocation reflects Nadia's priorities: voter contact programs receive the largest share because they have the most direct effect on the election outcome; polling is a significant investment because internal data quality is non-negotiable; technology is a necessary overhead; and staff are lean (just four people managing a significant operation) because the campaign's resources are tight.

Staff vs. Contractor Decisions

The analytics team's small size reflects a deliberate decision to minimize permanent staff and maximize contractors for specific deliverables. The direct mail program is managed by a direct mail vendor (included in the mail line item) rather than hiring an in-house mail coordinator. The digital advertising is managed through a digital consultancy (included in the digital line item). Voter file data appends are purchased from a vendor rather than built in-house.

This contractor model has advantages and disadvantages. Advantages: lower overhead, access to specialized expertise, easier to scale up and down, and contractors who serve multiple campaigns often have better knowledge of what's working across the industry. Disadvantages: less control over contractor performance, coordination overhead, and the risk that contractors prioritize their other clients' needs over this campaign's urgent needs.

Nadia's rule for the contractor vs. staff decision: "If a function requires day-to-day judgment about campaign strategy, it needs to be in-house. If it's a defined deliverable that can be specified in a contract and measured on delivery, it can be a contractor. Mail production is a contractor. Analytics planning is in-house."

Technology Stack: VAN, Digital Tools, Reporting Tools

VAN (Voter Activation Network): The campaign's primary operational platform. VAN stores the voter file, manages the walk list generation and assignment, records field contacts, manages the phone banking operation, and tracks volunteer information. It is the single source of truth for all field program data.

TargetSmart / L2: Voter file data vendors that provide commercial data appends and model score updates. The campaign purchases a monthly voter file refresh to capture new registrations, address changes, and updated vote history as early voting begins.

Digital advertising platforms: The campaign runs display and video advertising through a combination of Meta (Facebook/Instagram), Google Display Network, programmatic display platforms, and Hulu/streaming video. The voter file is matched to these platforms using the campaign's digital targeting vendor, which maintains current opt-in audience lists matched from the voter file.

Reporting infrastructure: The analytics team uses a combination of Python (for voter file analysis, universe building, and model evaluation) and Tableau (for dashboard visualization accessible to non-technical campaign staff). Weekly briefs are produced in a standardized Google Docs template. The voter contact dashboard (from Chapter 33) is the primary operational monitoring tool.

Making the Case for Analytics Investment to the Campaign Manager

Campaign manager Renata Diaz is not a skeptic of analytics — she hired Nadia, after all — but she is a practitioner who needs to justify every budget allocation to the campaign's finance team. When the final 60-day budget is finalized, she asks Nadia to justify the analytics spend.

Nadia's case rests on four arguments:

Argument 1 — The voter file is the campaign's most valuable asset. The 3.7 million voter records, enriched with model scores and commercial data, represent years of prior cycles' investment by the Democratic Party and aligned organizations. The analytics budget maintains and extends this asset. Failing to use it effectively means wasting prior investment.

Argument 2 — Data-driven resource allocation prevents waste. Without analytics, the campaign would distribute field resources according to gut instinct or political relationships (give more to counties where the organizer is a personal connection). Analytics shows where the marginal dollar of field investment has the highest expected return. Conservative estimates suggest that data-driven geographic allocation improves the efficiency of the field program by 15-25 percent — a significant return on the analytics staff cost.

Argument 3 — Internal polling is insurance. The cost of getting the strategy wrong in a 60-day stretch is far higher than the cost of internal surveys that catch strategic errors early enough to correct them. A $28,000 county-level survey that reveals the campaign is over-investing in a county it's already winning (and should be investing more in a county it can flip) could redirect hundreds of thousands of dollars in field spending to higher-impact activities.

Argument 4 — Post-election accountability builds institutional value. The postmortem documentation creates lasting organizational learning for the Democratic Party in the state. The models, methods, and lessons from this campaign will inform the next campaign. The analytics investment is not purely sunk cost — it builds infrastructure that compounds across cycles.

Section 8: Ethics and Equity Review

Ethics Checklist: Is Every Component of This Plan Ethically Sound?

Before the analytics plan goes to Renata for final approval, Nadia requires a documented ethics review. This is not a formality — it is a substantive check on whether every component of the plan can be defended as consistent with the campaign's stated values and with the norms of democratic practice.

The ethics checklist works through each program component:

Voter file and targeting data

Question: Is all data used in the plan obtained legally and ethically? Assessment: The voter file is obtained through the state Democratic Party's standard vendor contract. Commercial data appends are purchased from vendors who comply with applicable privacy regulations. No data is obtained through hacking, theft, or misrepresentation. Rating: Clear.

Question: Are there data uses that are legal but ethically problematic? Assessment: The voter file contains sensitive inferred data (income, health indicators, etc.) that voters did not explicitly consent to share with the campaign. The analytics plan's ethics constraints (Section 3) prohibit the most sensitive uses. The remaining uses (demographic targeting, behavior-based segmentation) are within the norms of democratic campaign practice. Rating: Acceptable with constraints documented.

Field program

Question: Are canvassers trained to represent the campaign accurately, without deception? Assessment: The canvassing training manual explicitly prohibits misrepresentation of Whitfield's positions, fabrication of quotes or claims, and deceptive presentation of the campaign's identity. Rating: Clear.

Question: Does the canvassing program respect voters' privacy? Assessment: Canvassers knock on doors — publicly accessible spaces. They do not access private property, do not photograph voters, and do not record conversations without consent. Rating: Clear.

Digital advertising

Question: Are all digital ads clearly identified as political advertising? Assessment: Federal law requires political advertising on major digital platforms to include "Paid for by" disclosures. The campaign complies with all disclosure requirements. Rating: Clear.

Question: Are any digital targeting techniques deceptive? Assessment: The campaign does not use deceptive ad identities, false endorsement claims, or misleading URLs. Rating: Clear.

Polling

Question: Are survey respondents informed of the survey's purpose? Assessment: Meridian's informed consent protocol discloses that the survey is a political opinion poll and identifies the research firm (though not the campaign client, which is standard practice for internal polls). Respondents are informed of how long the survey will take and can decline at any point. Rating: Clear.

Question: Are polling results used to manipulate rather than inform? Assessment: Internal poll results are used for strategy decisions, not for push-polling (calling voters under the guise of a poll to deliver negative messages). Rating: Clear.

Targeting decisions

Question: Does the targeting plan comply with anti-discrimination law? Assessment: The analytics plan does not use race or religion as targeting criteria in digital advertising (which would violate platform policies and potentially fair housing law for digital ad targeting). Targeting uses behavioral and demographic variables that correlate with voting behavior without constituting discriminatory targeting. Rating: Clear with monitoring.

Equity Review: Does the Targeting Plan Systematically De-prioritize Any Community?

The equity review is separate from the ethics review because equity questions are not primarily about legality but about fairness and access. A targeting plan can be legally and ethically sound while still systematically disadvantaging certain communities — for example, by concentrating high-quality contact programs in affluent, English-speaking areas while delivering lower-quality outreach to low-income communities of color.

The equity review examines the analytics plan along four dimensions:

Dimension 1 — Resource allocation by community

The analytics plan allocates the most intensive (and expensive) contacts — canvassing and relational organizing — to Tier 1 universe voters in high-density urban areas. These areas are disproportionately communities of color and are also the most cost-efficient canvassing targets (short distances between doors, high universe density). The resource allocation, in this case, aligns with both efficiency and equity — the communities receiving the most intensive outreach are both the most cost-efficient to reach and communities whose political engagement the campaign has the highest interest in supporting.

Where the review identifies a potential equity concern: rural Garza voters (a smaller universe, spread across large geographic areas, disproportionately Latino farmworkers in the state's agricultural interior) receive primarily mail outreach because in-person canvassing is not cost-efficient there. Mail is less effective than canvassing. These voters receive lower-quality outreach partly because of where they live. The analytics plan flags this and recommends that the community-based partner organizations in agricultural communities (which the campaign has relationships with) be resourced to conduct peer GOTV outreach — a relational model that doesn't require campaign staff to cover the geographic distance.

Dimension 2 — Language access

The campaign's Spanish-language program (Spanish-language mail, Spanish-language phone banking, Spanish-language digital advertising) covers the GOTV and persuasion universes for Spanish-surname voters. The analytics plan assesses the coverage: approximately 78 percent of Spanish-speaking Garza targets receive at least one Spanish-language contact. The remaining 22 percent are in geographic areas not covered by the Spanish-language program due to insufficient Spanish-speaking volunteer capacity.

The equity review recommends: prioritize Spanish-speaking volunteer recruitment in the three counties with the largest uncovered Spanish-speaking GOTV universe.

Dimension 3 — Young voter access

Young voters (18-29) face structural barriers to voting — unfamiliarity with processes, residential mobility that creates registration problems, work schedules that conflict with voting hours. The analytics plan's young voter program (text messaging, digital advertising, campus organizing) is assessed for accessibility: Does the program reach young voters where they are? Does it provide actionable information (registration deadlines, polling locations, early voting options) rather than just motivational messaging?

The review identifies a gap: the campus-based organizing program covers three universities but not the community college system, which has disproportionately higher enrollment of students of color, first-generation college students, and working-class young adults. The analytics plan adds community college outreach to the campus organizing budget.

Dimension 4 — Voters with disabilities and elderly voters

Voters with disabilities and elderly voters may face barriers to both in-person voting and responding to standard contact programs. The analytics plan notes that the mail program's accessibility (large-print option for voters identified with visual impairment flags in commercial data) is an existing strength, but the canvassing program does not have a systematic accommodation protocol for voters with mobility limitations. The equity review recommends adding an accessibility accommodation protocol to canvassing training.

Nadia's Test: "Would I Be Comfortable If This Plan Were Public?"

Nadia has a final heuristic for ethical review that she applies to everything: "If the New York Times published a front-page story about exactly what we're doing and why, would I be comfortable with it?"

She walks through the plan's major components:

"We're building a voter file universe using public voting records, demographic data, and vendor-provided model scores. We're using those universes to target outreach — canvassing, phone, mail, digital — with messages matched to voter priorities. Would I be comfortable with that on the front page? Yes. Campaigns have done this for decades. The targeting is based on political behavior and demographic characteristics, not exploitative personal data."

"We're running split tests on our mail and canvassing scripts to see which versions are more effective. Would I be comfortable with that? Yes. This is how we improve. Randomly testing different messages and measuring outcomes is good science."

"We're purchasing voter file data appends that include consumer behavior indicators. We're using those to refine model scores but not to target individual voters based on inferred health or financial stress. Would I be comfortable with the data purchases? Yes. Would I be comfortable if we were using inferred health data to target individual voters with scare messages about their personal medical situation? No. And we're not doing that."

"We're doing internal polling, including message testing. We're using the poll results to refine our strategy, not to release favorable numbers to suppress opposition enthusiasm. Would I be comfortable with how we're using polling? Yes."

The one area where Nadia pauses: the digital advertising's voter file-to-platform matching. "The matching process means we're providing voter file data to advertising platforms to enable targeting. That data isn't transmitted in a way that gives the platform access to individual voter records — it's a hash-matching process — but the principle of voter data being shared with commercial platforms for advertising purposes makes me uncomfortable. I want us to be aware of it and to monitor platform policy changes that might affect how that data is used."

She doesn't eliminate this component — digital advertising matched to the voter file is standard practice and the campaign's digital program depends on it — but she documents the concern and commits to reviewing it if platform policies change during the campaign.

Section 9: Final Deliverable — The Complete Plan

How to Present the Analytics Plan to Campaign Leadership

The analytics plan document you have built across Sections 2-8 is approximately 35-40 pages of detailed analysis, tables, and methodology. This is not what Renata Diaz will read in the campaign's weekly leadership meeting.

Campaign leadership needs a compressed, decision-focused presentation of the analytics plan. Nadia has a template for this, and it forms the basis of the executive summary format you'll need to master.

The one-page overview: This is what campaign leadership sees first. It contains:

1. Strategic situation (three sentences): What the data says about the current state of the race — horse race number, key dynamics, opportunities and risks.

2. Universe summary (one table): Persuasion universe (total, Tier 1, Tier 2, Tier 3), GOTV universe (total, priority breakdown), by county. No more than 10 rows; additional detail in appendix.

3. Contact program projections (one table): For each program (canvassing, phone, mail, digital, text), projected total contacts over 60 days, projected cost, and projected universe penetration rate.

4. Key decisions (bullet list): The three to five decisions that require campaign leadership input. Each presented as a specific choice with the expected outcome of each option and Nadia's recommendation.

5. Risk flags (bullet list): The two or three things the analytics team is most concerned about and monitoring.

6. Bottom line: One sentence. "Based on current data, with full plan execution, we project Garza winning with approximately [X] to [X+6] point margin. The primary risk is [specific factor]."

The one-page overview is designed to be read in three minutes and to enable a 15-minute decision conversation with campaign leadership. The 35-page detailed plan is attached for anyone who wants the full methodology — and Nadia expects that some people will read it.

Executive Summary Format

The executive summary (two to three pages) bridges the one-page overview and the full plan. It tells the story of the analytics plan in a form accessible to senior campaign staff who need more than the overview but don't need the full methodology.

The executive summary structure:

Paragraph 1 — Why this plan: The analytics plan is the campaign's guide to deploying resources efficiently in the final 60 days. Every recommendation is grounded in data from the voter file, internal polling, and the campaign's own contact program results to date.

Paragraph 2 — The state of the race: What the current data shows about the electoral landscape. Where Garza is strong, where she's competitive, where she's behind.

Paragraph 3 — The target universes: Who the campaign is targeting and why. High-level overview of persuasion and GOTV universe logic without the technical detail.

Paragraph 4 — The contact program: How the campaign will reach target voters. Channel mix rationale and timeline overview.

Paragraph 5 — The measurement framework: How the campaign will know if the plan is working. Key metrics and the weekly review process.

Paragraph 6 — Budget summary: Where the money is going and why.

Paragraph 7 — Ethics and equity: A brief statement that the plan has been reviewed against the campaign's ethics principles and equity commitments, with a pointer to the full review documentation.

Paragraph 8 — The ask: What campaign leadership needs to approve. (In most cases, this is the full plan and budget. Sometimes it's a specific decision within the plan — a budget reallocation, a new program, a strategic shift.)

The Analytics Director's Sign-off Process

When the analytics plan is complete, Nadia signs off on it by doing four things:

First, she reads it cover to cover, checking for internal consistency. Do the universe sizes add up? Do the budget numbers in Section 7 match the program descriptions in Sections 4 and 5? Are the KPIs in Section 6 measurable given the data collection described in the program sections?

Second, she stress-tests the key assumptions. The plan assumes a canvassing contact rate of 25-35 percent. What happens if the actual contact rate is only 18 percent? The plan projects a 70 percent penetration of the Tier 1 GOTV universe. What happens if it's only 55 percent? She runs through the key quantitative assumptions and documents what she'd do if they proved wrong.

Third, she discusses the plan with you — her analytics associate — and asks you to challenge her. "Tell me where you think I'm wrong. Tell me what I've missed." This is not performative — she genuinely wants to hear objections. The plan is better for having been challenged.

Fourth, she submits it to Renata with a cover note that says, explicitly: "This is my best analysis based on current data. It will change as we get new data. I'll flag material changes in the weekly brief. I'm confident in the process even when I'm not certain about specific outcomes."

That last sentence is worth pausing on. Nadia is confident in the process. The process — the systematic universe building, the targeted contact programs, the regular measurement, the willingness to update based on new evidence, the ethics and equity review — is something she has built carefully and can stand behind. The outcomes are uncertain. That uncertainty is honest. An analytics director who claims to be certain about outcomes is either lying or doesn't understand what their models can and cannot tell them.

You have spent 60 days building this plan. On Election Day, you will find out if you got enough of it right. Whatever the result, you will write the postmortem.

A Final Word

This capstone asked you to inhabit a role: Nadia's analytics associate, building a real plan for a real-feeling campaign. You have worked through voter universe construction, targeting strategy, field program design, polling, measurement, budget allocation, and ethics review. You have made choices — some with clear right answers, others genuinely debatable — and documented your reasoning.

That documentation is the most important thing you built.

Political analytics, at its best, is not about finding clever shortcuts or deploying sophisticated models to achieve ends that wouldn't survive ethical scrutiny. It is about the systematic, honest, humble application of data and analytical tools to the irreducibly human project of democratic participation — helping people connect their values to their votes, helping campaigns find the voters most likely to respond to their message, and ensuring that the resources available to a campaign are deployed in ways that can be defended to the public.

Nadia could have built this plan without you. She's done it before. What you gave her was a structured process, a documented rationale, and a second set of eyes on every major decision. In analytics — as in science, journalism, and every other discipline that deals honestly with uncertainty — that is exactly what the work requires.

The race is over. Whoever won, the work continues.

This concludes Part IX: Capstone Projects. Appendices A through I provide supplementary materials including research methods reference, Python toolkit, data sources guide, historical timeline, and a complete glossary of terms introduced throughout the textbook.