An intersection in the Central Havana neighborhood. (Sarah L. Voisin/The Washington Post)

How do you conduct a reliable public opinion poll in a closely monitored society where political dissent is strictly repressed?

That was the significant hurdle for the Univision Noticias/Fusion poll of Cubans, reported in collaboration with The Washington Post. The new poll offers an extremely rare and in-depth nationally representative examination of opinions of adults living in Cuba.

This survey, conducted by the Miami-based research firm Bendixen & Amandi International (B&A), was designed and executed without the authorization of the Cuban government. Interviews were conducted in person among a random national sample of 1,200 Cuban adults from March 17 to 27. The overall margin of sampling error for the poll is plus or minus three percentage points at the 95 percent confidence level.

Interviews were conducted by Cuban residents personally trained by senior B&A survey researchers. Prior to the main survey, a seven-day pilot study was conducted to ensure the poll’s feasibility given potential concerns about confidentiality and privacy.

Before taking the survey, interviewers assured respondents that individual answers would remain confidential and that their answers would be analyzed only in the aggregate. Among all the households sampled for an interview, 39 percent completed one.

To complete personal interviews using random selection techniques, a multi-stage process of respondent selection was employed.

Researchers began by dividing the population by province, urbanization level and population density. Within each segment they chose primary sampling locations, with more-populated areas being given a proportionately greater chance of being selected. In total, 120 sampling areas were selected throughout Cuba’s 14 provinces (excluding the small Isla de la Juventud).

At each sampling point, trained interviewers traveled to a randomly chosen household. Once the first selected household was completed (or unavailable), interviewers selected additional households using a random route method, taking alternate right and left turns and selecting households at predetermined intervals.

Finally, respondents were selected within each sampled household using a random selection grid (or Kish grid). At least three attempts were made to reach the selected individual, after which interviewers moved to the next house according to their random route procedures.

Interviewers recorded responses with handheld electronic devices. The responses then were downloaded to laptops and delivered to a server outside the country. Data was tabulated by B&A, and original responses on devices were deleted.

The demographic profile of survey respondents closely reflected the known population distribution based on race, gender and region within Cuba. Poll respondents were younger than the Cuban population more broadly. Data was not weighted.

Update: An earlier version of this article gave incomplete information about the procedures used when poll takers were not able to interview a sampled respondent. It has been updated.

Scott Clement contributed to this report.

Read more:

The political demography of U.S.-Cuba relations

Four reasons an opening to Cuba is long overdue

Today’s coverage from Post correspondents around the world