Case Study 9-1: Meridian's Multi-Mode Crisis

The Situation

It is day three of a five-day field window for Meridian Research Group's first major poll of the Garza-Whitfield Senate race. The client — a national Democratic Senate campaign committee — has paid for a 700-likely-voter sample with a ±3.7% margin of error. Results are needed by Friday morning for a Sunday release coinciding with a campaign finance fundraising push.

Trish McGovern opens the morning dashboard to find the field operation is in trouble.

Current completion counts by mode (target in parentheses):

The problems are not merely about numbers. The demographic breakdown of completed interviews shows serious imbalances:

• Hispanic respondents: 6.2% of sample (target: 14.8% per voter file)
• Rural county respondents: 8.1% of sample (target: 19.3%)
• Respondents aged 18–34: 9.4% of sample (target: 17.6%)
• Respondents aged 65+: 31.7% of sample (target: 21.2%)

Two days remain in the field window.

Complications

Complication 1 — The CATI problem: The telephone facility Meridian uses reports that their Monday-Wednesday evening call center staffing has been reduced by 30% due to a state health emergency affecting their workforce. They can restore capacity Thursday evening, which is the final field day.

Complication 2 — Hispanic outreach: The CATI facility only has Spanish-language capacity for daytime hours (9am–3pm). The questionnaire is 22 minutes long. Daytime response rates for working-age Hispanic voters are extremely low. The Spanish-language component has completed only 19 interviews in three days against a target of 40.

Complication 3 — Rural gap: The mail/web ABS component, which was specifically designed to capture rural respondents with limited online panel presence, is returning at a 44/100 rate with two days remaining. Follow-up mailings take 5–7 days. There is no feasible mail remedy within the field window.

Complication 4 — Client pressure: The campaign committee's analytics director has called Trish directly to ask whether she can "bump the online panel" to fill the gaps and deliver the 700-interview target on schedule. She frames it as "just use your opt-in backup for the rural and Hispanic gaps."

The Decision Point

Trish must decide how to respond to each complication. She lays out the options in a team meeting with Carlos and the CATI facility manager:

Option A: Accept reduced sample, deliver 562 completed interviews. Report actual sample size, adjust the margin of error to ±4.1%, and deliver on schedule with transparent documentation of the shortfall.

Option B: Extend the field window by two days. Add Saturday–Sunday calling and online completion time. Restores full CATI capacity, allows more rural/Hispanic outreach. Requires client approval and delays the Sunday release.

Option C: Fill gaps with opt-in online panel. Purchase 138 additional completes from Meridian's opt-in backup vendor. Meets the 700-interview target and Friday deadline. Introduces non-probability completions that would not meet AAPOR probability-sample standards and would need careful disclosure.

Option D: Hybrid extension. Add 48 hours to the online probability panel and CATI components while accepting a smaller sample (600 interviews, ±4.0% MOE). Use intensive CATI Spanish-language calling during Thursday daytime hours. No opt-in fill. Document rural gap as a limitation.

Discussion Questions

Question 1: What are the methodological consequences of each option, specifically regarding: (a) the stated margin of error, (b) representation of Hispanic and rural subgroups, and (c) the validity of comparing results to future polls using a different mode composition?

Question 2: The client's request to "bump the opt-in online" to fill gaps is not unusual in commercial polling. What would Meridian's AAPOR reporting obligations be if Option C were chosen? Draft the methodology disclosure language that would be required.

Question 3: Using the nonresponse bias formula discussed in Section 9.3.1, estimate the potential directional bias in the topline if Hispanic and rural voters are systematically underrepresented at the rates shown in the current sample. You may assume: Hispanic voters support Garza at 72%, rural voters support Whitfield at 63%, and the overall sample so far shows Garza +4.

Question 4: Trish believes Option B is the right methodological choice but will face client resistance. Draft the email she should send to the analytics director explaining why a 2-day extension serves the client's interests better than rushing with a compromised sample.

Question 5: If Trish chooses Option D and the rural gap remains, how should this limitation be reported in the poll's public release? Write the specific language that should appear in the methodology footnote.

What Meridian Did

Trish chose a modified version of Option D. She extended the online probability panel and CATI components by 36 hours, concentrated all available Spanish-language CATI capacity on Thursday daytime calls (completing 28 additional Hispanic respondent interviews), and accepted a final sample of 618 likely voters (MOE ±3.9%). The rural underrepresentation was documented in the methodology note with a flag that rural subgroup estimates should be interpreted cautiously.

The client released the poll on Monday rather than Sunday, losing the fundraising window they had planned but gaining a methodologically defensible result. When a rival pollster released a same-week poll showing a 3-point gap in the opposite direction, Meridian's clean methodology disclosure gave the campaign committee confidence in their own numbers.

Trish's note to Carlos after the project closed: "You'll get clients who think data is like clay — you can mold it into the shape you need. Our job is to explain why some shapes aren't clay. This one we got right."

Learning Points

1. Multi-mode operations require contingency planning — no single mode is reliable enough to absorb all operational problems.
2. Demographic imbalances create substantive, not merely statistical, problems when underrepresented groups hold distinctive political views.
3. Client pressure for speed is almost always in tension with methodological soundness. The analyst's role includes making the trade-off explicit.
4. Opt-in fill-in of probability sample gaps violates sample design in ways that cannot be fully corrected through weighting.
5. Transparent disclosure of limitations is not a failure — it is the mechanism through which users of poll data can apply appropriate interpretive caution.