The venerable Gallup Organization no longer conducts polls of U.S. presidential races. It dropped out after mistakenly estimating that Mitt Romney narrowly led President Barack Obama at the close of the 2012 campaign.
Even though Gallup has quit election polling, a reservoir of its reporting remains available online; it’s a useful, even revealing resource about campaigns past. For example, in 2008, Gallup posted an article online that noted polls of July are not always reliable indicators of outcomes in November.

Miscalling the 2012 election
It is a relevant reminder, even now as polls signal President Donald Trump’s reelection seems a long shot in his unfolding race against gaffe-prone Joe Biden.
The winners of July who lost in November have been several over the years, and go beyond the cases of Thomas E. Dewey in 1948 and Hillary Clinton in 2016. Among the others in Gallup’s polling were:
- John Kerry, who in July 2004 held a seven-point lead over George W. Bush. Kerry lost by 2 percentage points.
- George Bush, who in July 2000 had two-point lead over Al Gore. Bush lost the popular vote by 0.5 percentage points — possibly because of the 11th hour disclosure that he had been arrested years before for DUI and had pleaded guilty to the offense.
- Bush’s father, George H. W. Bush, who led Bill Clinton by 7 percentage points in July 1992. Bush lost in November by 5.5 percentage points — a 12.5-point turnaround.
- Michael Dukakis, who led George H. W. Bush by 6 percentage points in July 1988. Dukakis lost the election by 8 points — a 14-point turnaround.
- Hubert H. Humphrey, who led Richard Nixon by 5 percent points in July 1968, and lost by 1 point in November.
Other races have demonstrated how deceptive the polls of July can be.
In mid-July 1976, Jimmy Carter opened a 33-point lead over Gerald Ford; Carter won by two points in November. (And Gallup’s final poll before the 1976 election showed Ford ahead by a point.)
John F. Kennedy led Nixon by 6 percentage points in July 1960, and barely hung on to win in November. And in July 1996, Bill Clinton led Bob Dole by 17 percentage points, a margin that was sliced by half on Election Day.
Gallup’s poll in July 1936 gave Alf Landon an electoral vote lead
Even at the dawn of modern opinion polling, summertime polls proved misleading. George Gallup, then a polling novice, reported in July 1936 that Republican Alf Landon was narrowly ahead of President Franklin D. Roosevelt in electoral votes. Landon led, Gallup reported, in 21 states with 272 electoral votes — six more than votes than what was needed at the time to win the presidency.
As I discuss in Lost in a Gallup, my forthcoming book about polling failure in presidential elections, Gallup’s poll in July 1936 “corresponded to a notion that the summer that the election would be close.” Even the best-known poll of that time — the Literary Digest‘s mass mail-in survey — indicated that Landon would win.
Easily.
He didn’t. Landon carried just two states — Maine and Vermont — as Roosevelt rolled up a crushing victory in November 1936. Landon lost even his home state, Kansas, where he was governor.
So what do these cases tell us?

A cartoonist’s view of the ’48 election
They tell us, obviously, that voter sentiment is not static.
Views about candidates can and do shift as Election Day nears — even in races that seem to be near-blowouts in summertime. Consider, for example, the 1948 campaign, when President Harry S. Truman trailed Dewey by 11 percentage points in Gallup’s late-July survey. Elmo Roper’s polling in July 1948 showed Truman behind by nearly 15 percentage points.
The cases of poll-winners of July also underscore that campaigns — particularly focused, relentless, and disciplined campaigns of the kind Truman ran to win reelection in 1948 — can make a difference.
Campaigns can tell voters much about candidates. Landon, Dewey, Dukakis, and Kerry all ran dismal, uninspiring campaigns. And they effectively were defined by their opponents.
The winners of July also remind us that summertime polls are not prophesies, that they may have limited predictive value. As I have noted, presidential races have been known to turn on late-in-the-campaign developments — which may have been the case in 2016, when the FBI announced in late October it had reopened its investigation into Hillary Clinton’s private email server.
That development might have shifted enough votes in key states to Trump, enough for him to win an Electoral College majority.