Dear Kate,

Thank you for sending the Embold Research methodology statement. I have reviewed it carefully.

The statement answers some of my questions, but several important methodological details remain undisclosed. Rather than asking generally for additional information, I would like to identify the specific items I am requesting so that each can be answered separately.

The methodology statement says that 815 likely Rhode Island voters were surveyed October 5–8, with 452 respondents recruited through targeted Facebook/Instagram/web advertising and 363 through text messages to voter-file telephone numbers. It also states that Embold dynamically adjusted advertising budgets and texting during fielding to increase recruitment of groups it considered underrepresented.

Could you please provide the following:

1. Meta/Facebook/Instagram targeting criteria

What specific targeting parameters were used for the Facebook and Instagram advertisements?

Please include, to the extent applicable:
- age ranges;
- gender;
- geographic targeting;
- political or voter-related targeting;
- interests or behavioral characteristics;
- custom audiences;
- lookalike audiences;
- voter-file-derived audiences;
- exclusions; and
- any other audience-selection criteria used.

2. Copies of or records for the advertisements

Please provide the actual advertisements used, or links/identifiers sufficient to locate them in Meta's advertising records, together with the dates each advertisement ran.

3. Advertising expenditures by target group

The statement says Embold raised or lowered advertising budgets depending upon which groups were underrepresented or overrepresented.

Please provide the amount spent on each audience or target group and any record showing how those expenditures changed during the October 5–8 field period.

4. Recruitment results by target group

How many completed respondents came from each advertising audience or demographic target?

In other words, I am asking not merely for the total of 452 ad-recruited respondents, but for the underlying breakdown showing which targeting efforts produced those respondents.

5. Day-by-day changes during the survey

Because Embold states that it changed advertising expenditures and texting during the four-day field period in response to which demographic groups were overrepresented or underrepresented, please provide a day-by-day breakdown for October 5, 6, 7 and 8.

For each day, please provide:
- the number of completed responses received;
- the number recruited through Meta/web advertising;
- the number recruited through text messaging;
- the demographic composition of respondents received that day;
- which demographic or other groups were being specifically targeted that day;
- which Meta audiences or targeting parameters were active;
- any increases or decreases in advertising expenditures by audience;
- any changes in the groups selected for additional text-message recruitment; and
- the unweighted ballot results received that day.

For the Attorney General race specifically, please provide the day-by-day number and percentage supporting Alan Gordon, Kim Ahern and Not Sure.

I would also like to know whether the Attorney General results changed as Embold altered its recruitment strategy during the four days of fielding.

For example, as additional demographic groups were recruited through increasingly targeted Facebook/Instagram advertising or text outreach, did the Gordon/Ahern percentages move? If so, by how much and in which direction?

This information seems particularly important because the methodology describes active intervention in the composition of the incoming sample while responses were still being collected.

6. Definition of the survey's “sample universe”

The statement says text messages were sent to individuals who qualified for the survey's sample universe based on L2 voter-file data.

What exact criteria determined whether a voter qualified for that universe?

7. Likely-voter criteria

How did Embold determine who constituted a “likely 2026 voter” before or during recruitment?

Please provide the variables, scores, voter-history requirements, modeling criteria, or other rules used.

8. Text-message sampling

For the Echo19 recruitment:
- how many telephone numbers were selected;
- how many texts were sent;
- how those numbers were selected from the L2 voter file;
- whether different voter groups were sampled at different rates;
- how many people clicked the survey link;
- how many began the survey; and
- how many completed it.

9. Unweighted sample composition

Please provide the unweighted counts and percentages of the 815 respondents before post-stratification for each variable used in weighting:
- age;
- gender;
- race/ethnicity;
- education;
- region; and
- reported 2024 presidential vote.

10. Party composition before weighting

Because the questionnaire asked respondents both how they politically identify and how they are registered, please provide the unweighted number and percentage identifying as:
- Democrat;
- Republican;
- Independent/Other;

and the unweighted number and percentage registered as:
- Democratic;
- Republican;
- third party; and
- unaffiliated.

11. Weighting targets

Please provide the numerical population targets used for every category of every weighting variable.

For example, rather than simply stating that the sample was weighted by age, gender, race/ethnicity, education, 2024 presidential vote and region, I am requesting the actual percentage target assigned to each category.

12. Individual or category weighting factors

Please provide the weighting coefficients applied to respondents, or alternatively sufficient information to reproduce the weighting.

At minimum, please provide:
- minimum weight;
- maximum weight;
- mean weight;
- median weight; and
- distribution of weights.

13. Design effect

The methodology statement says the reported 3.6% modeled margin of error uses an effective sample size adjusted for the design effect of weighting.

What was the calculated design effect?

14. Effective sample size

What was the effective sample size after application of the weighting, using the Kish approximation referenced in the methodology statement?

15. Unweighted ballot results

Please provide the unweighted results for each ballot question, particularly the Attorney General race, before post-stratification weighting was applied.

16. Effect of weighting on the Attorney General result

Please provide the Attorney General result both before and after weighting, including the weighted and unweighted number of respondents supporting:
- Alan Gordon;
- Kim Ahern;
- Not Sure; and
- Would Not Vote, where applicable.

17. Recruitment-source results

Please provide the Attorney General result separately for:
- the 452 respondents recruited through Meta/web advertising; and
- the 363 respondents recruited by text message.

18. Survey dispositions

Please provide the total number of people who:
- entered the survey;
- were terminated because they were not registered;
- were terminated based upon likelihood of voting;
- abandoned the survey;
- completed it; and
- were otherwise excluded from the reported sample.

19. Quality-control exclusions

Were any completed or partially completed responses excluded because of duplicate responses, speed, inconsistent answers, IP/device information, suspected bots, inaccurate demographic information, or other quality-control criteria? If so, how many and under what rules?

20. Selection of Embold Research

How did the Pell Center select Embold Research for this project?

Was there a competitive proposal process, prior relationship, recommendation, or other selection process?

21. Information considered when selecting Embold

Before selecting Embold, did the Pell Center review the firm's ownership, principals, previous political clients, partisan campaign work, or other political affiliations or professional relationships? If so, what information was considered?

I recognize that this is a detailed request. The reason for the detail is that the published methodology describes two separate stages capable of materially changing the composition of the final poll: first, dynamic recruitment in which advertising expenditures and texting were adjusted to obtain particular groups, and second, post-stratification weighting of the resulting respondents.

Without the underlying recruitment and weighting information, it is difficult for an outside reader to determine how much of the reported result arose from the respondents who actually participated and how much arose from the sampling, targeting and weighting decisions made by the pollster.

There is an additional reason I believe transparency is particularly important at this moment. Two widely reported fabricated-poll scandals have now occurred during this election cycle, with the most recent involving Wedgewood Polls breaking publicly this week. That episode has generated renewed national discussion about the importance of access to underlying methodology and data when political polling is presented to the public.

I am not suggesting that the Pell Center or Embold engaged in such conduct. Rather, these events demonstrate precisely why meaningful transparency is important for legitimate polling organizations.

The Pell name is highly respected, particularly here in Rhode Island. In a small and close-knit state, a poll released under the name of an institution with that standing can carry substantial public and media influence. For that reason, I believe complete transparency concerning how respondents were recruited, how targeted advertising altered the sample during fielding, and how weighting changed the eventual results is of paramount importance.

If any of the requested information does not exist, was not retained, or will not be provided, please identify the particular item and say so. That would also answer the question.

Thank you for your attention to this request.

Prayerfully,
Rev. Dr. Anne Armstrong