In 2012, national polls in October suggested the presidential race was a virtual tie. The Real Clear Politics polling average gave Barack Obama a slight 0.7 point lead over Mitt Romney, but he actually won by almost 4 points. Of the final 11 national polls released in 2012, as reported on Real Clear Politics, 7 were a tie or had Romney ahead, while only 4 had Obama ahead.

Why were so many of the polls wrong? In part, because they failed to capture how minorities would vote. Unfortunately, some pollsters may be making the same mistakes in 2016 — and thereby underestimating Hillary Clinton’s lead in the polls.

In 2012, many polls underestimated how many minorities would vote and how many would vote for Obama. For example, a Politico poll released the morning of Election Day said the race was tied at 47 percent each for Obama and Romney.  The poll said that 62 percent of Latinos supported Obama, while the exit polls reported 71 percent, and Latino Decisions reported 75 percent. Among the “another race” category, which is mostly comprised of Asian Americans, Politico reported that 47 percent supported Obama, while the exit polls reported 73 percent, and an Asian American Decisions exit poll reported 72 percent.

And Politico was not alone. A Monmouth/Survey USA poll, which had Romney leading by 3 points, suggested that Obama would barely win Latinos, 48 percent to 42 percent.

This problem was known before election day. In the fall of 2012, Mark Blumenthal asked “Is The Gallup Poll Favoring Mitt Romney By Undersampling Minority Voters?” which came after a series of blog posts by Alan Abramowitz, one of which asked “Is Gallup Heading for Another Big Miss?”  As Nate Cohn has recently pointed out, it is difficult to know the “correct” percent of voters that are white vs. non-white. Nevertheless, many 2012 polls underestimated Obama’s share of the vote by under-representing minorities’ share of the electorate and underestimating their support for Obama.

Now, in 2016, it looks like many pollsters didn’t learn much from 2012.

Several polls suffer from flaws in how they sample Latinos. While large bilingual polls of Latino voters from outlets such as Latino Decisions and Univision/Washington Post have reported very little support for Donald Trump, other national polls that interview Latinos only in English show that 29 percent to 37 percent of Latinos will support Trump — better even than Romney fared. These polls also show Trump doing as well or better among Asian-Americans, compared to Romney. Some polls even estimate that 20 percent to 25 percent of blacks support Trump.

Here is one example from a Survey USA poll conducted on behalf of The Guardian, which gave Clinton a 3-point lead over Trump (39 percent vs. 36 percent). The sample of the poll was 74 percent white. However, a comprehensive analysis of census data and growth rates by Ruy Teixeira and William Frey estimates that less than 70 percentgof voters in the 2016 cycle will be white. If we adjust the racial composition of this poll to reflect the Teixeira and Frey’s estimates, Clinton’s margin grows to almost 5 points (see here).

The Survey USA poll is also arguably underestimating support for Hillary Clinton among three different groups of minority voters:

  • Asian-Americans. In this poll, Clinton leads 48-29 among Asian-Americans. However, a recent national survey of Asian American registered voters conducted in six different languages found that just 10 percent planned to vote for Donald Trump.
  • Latinos. In the Survey USA poll, Clinton had 45 percent of Latinos compared to 29 percent for Trump. However, a Latino Decisions poll of registered Latino voters that was conducted in both English and Spanish found that 87 percent of Latinos had an unfavorable view of Donald Trump and that only 11 percent planned to vote for him, compared to 76 percent for Clinton. A similar poll conducted by Univision and the Washington Post reported that 8 in 10 Hispanics viewed Trump negatively and that 16 percent planned to vote for him, compared to 73 percent for Clinton.
  • African-Americans. The Survey USA poll found that only 64 percent of African Americans said they will vote for Clinton compared to 17 percent for Trump. It seems hard to imagine that Clinton won’t get significantly more of the black vote. She is coming off a primary in which African-American voters buoyed her candidacy. And other polls, like a May NBC/Wall Street Journal poll, suggested a much bigger Clinton lead among blacks (88 percent vs. 9 percent for Trump).

When we adjust the Survey USA poll by substituting more realistic estimates for how minority voters will vote, Clinton’s margin grows even more. If Clinton does as well with minority voters as Obama did, then her lead in the poll would be 10 points (see here).

The point is not to pick on this particular poll, but to point out that the flaws in polling methodology in 2012 are reemerging now. This affects not only individual polls but also polling averages, which are aggregating polls with different samples and methodologies and which underestimated Obama’s vote share in 2012.

In 2016, the country is even more diverse. Pollsters need to take steps to more accurately estimate the political attitudes and behavior of black, Latino and Asian American voters. Polls that are conducted only in English and that do few or no callbacks to try contact hard-to-reach populations are not going to accurately reflect the American electorate. Polls that are conducted via online panels need to ensure that they are not under-representing minority voters with less formal education and lower incomes. Finally, pollsters should take steps to make sure each racial subgroup in their poll is weighted to match the American Community Survey’s estimates for that particular group, not just their national sample overall.

Polls that put more care and effort into understanding the minority electorate will likely be the ones who accurately estimate the national presidential vote in November.

Gabriel Sanchez is Professor of Political Science and Executive Director of the Robert Wood Johnson Foundation Center for Health Policy at the University of New Mexico.  He is also a Principal at the research and polling firm Latino Decisions.  The co-founders of Latino Decisions – Matt Barreto and Gary Segura – are currently consultants to the Clinton presidential campaign. However, Sanchez does not work on the Clinton campaign and instead is directing Latino Decisions non-candidate research projects in 2016.

Alan Abramowitz is the Alben W. Barkley Professor of Political Science at Emory University