Headlines in the media copied the detailed answers of a Marist Poll about New York City’s Mayor Zohran Mamdani’s first 100 days in office. Headlines indicate broad approval for the new Mayor.
The poll’s summary is that Mamdani is perceived as being likeable, hard working, caring, a good leader, representing all New York City residents, and a unifying force in the Big Apple. The poll states that the results are based on interviews conducted between March 26th and March 31st, 2026, using a 1,454 sample size, adding that “results for registered voters (n=1,247) are statistically significant within ±3.5 percentage points.” At closer range these numbers are statistical lies.
Following a stat showing that 48 percent of the interviewees approve of Mr. Mamdani’s performance, with 25% dissatisfied, the Marist compilation gives detailed numbers about how many Democrats and Republicans approved of his performance (63% and 25%, respectively), whereas “among those who are not enrolled in a party, 27% approve, 41% disapprove, and 31% are unsure.” These stats are followed by others such as those concerning boroughs, and a range of additional numbers, leading Marist Poll editors to the conclusion that “Residents Perceive Mamdani as Hard Working, Caring, Grasping NYC’s Problems, & Being a Good Leader.”
The page also says that “Adults 18 years of age and older residing in New York City were contacted through a multi-mode design: By phone using live interviewers, by text, or online. For full methodology and tables, click on the Survey Data button below.”
My guess is that no journalist summarizing the data (Home of the Marist Poll | Polls, Analysis, Learning, and More) clicked on that button. If they had, they would have found the following details. Subsequently they would have provided a far more skeptical take on the numbers:
“Adults 18 years of age and older residing in New York City were contacted through a multi-mode design: By phone using live interviewers (402), by text (714), or online (338). All potential respondents were screened for age and residence in New York City. The telephone sample was randomly selected from a stratified dual frame of landline and cell phone telephone numbers in the five boroughs of New York City. … For telephone surveys completed with live interviewers, household sampling was done by asking for the youngest male.” (italics added).
Why did the poll ask for this choice? How much bias does such choice imply for this poll? It’s impossible to know as the Marist detailed Appendix aggregated the 18-29 age group in one category. How many were closer to 18 – hardly having worked much in their lives, or earned much income at all (as with Mr. Mamdani himself) – is impossible to know. Another number in that Appendix mentions that 45% in the survey were not college graduates. It also mentions that 80% were “very liberal” and 86% in the sample voted for Mr. Mamdani, and that 46% had incomes less than $50,000. The partisan breakdown for this survey among registered voters is 62% Democrat, 15% Republican, and 23% non-enrolled. (Marist-Poll_NYC-NOS-and-Tables_202604021433.pdf).
From this, we can ask how reliable the answers are considering such strange decisions and aggregation? It’s impossible to assess the magnitude of errors due to such editing decisions – especially knowing that Mr. Mamdani won the mayoral vote with roughly 50%. Even this percentage reflects only the 2 million registered voters who showed up, not the 3 million who did not. Yet for any survey it would also be crucial to know how many of those contacted did not reply.
Add to the above that the Marist text mentions that “Coding of any open-ended responses was done by a single human coder.” This statement raises more questions about how Marist assesses the reliability of this survey. Who are their editors and how reliable is one coder deciding on how to categorize open-ended answers – whose numbers I could not find either.
The 3.5% margin of error the Marist Survey published is, as is well known, neither here nor there. It reflects a number derived from an “ideal statistical sampling model” when none of the specific issues raised above are present. Interested journalists and readers can look up MIT publications on the subject (such as https://news.mit.edu/2012/explained-margin-of-error-polls-1031), or Pew Research publication (https://www.pewresearch.org/short-reads/2016/09/08/understanding-the-margin-of-error-in-election-polls/), to just mention a few.
There is nothing new about the unreliability of political polling due to the fundamental errors mentioned. Researchers, such as the late political scientist Seymour Martin Lipset, advised decades ago against putting much stock in their reliability. However, he did not offer alternatives. Yet one now exists. Betting markets have the potential to outperform polling, and achieve far more, once their regulatory framework is settled.
Those markets will attract many independent players eager to put their money where their thoughts are, a process not dissimilar to what led to establishing futures markets in the past. Until then, perhaps media should be more aware of the old adage “lies, lies – and statistics.”
