This year’s polls were so bad I almost turned in my political-science license. Reputable surveys had given Biden an average lead of 8 percentage points; he won with 4. Such a 4-point miss (even bigger in House and Senate races) casts doubt on the scientific polling techniques George Gallup pioneered in the 1930s.

Prognosticators might have done better using COVID-19 cases. Republicans made going unmasked a political symbol and dismissed coronavirus as a hoax or not dangerous. Result: More Republicans got infected than Democrats, who tend to wear masks. The Associated Press found that 93% of the 376 counties with the fastest COVID growth went for Trump. Masks and votes correlate.

Improved survey accuracy — or at least warnings — is possible. The problem is not technical — sample size and selection, either random or stratified quotas — but the respondents, many of whom will not state their voting intentions. Most just hang up. In 1948 and 2020, most self-proclaimed “undecideds” voted, respectively, for Truman and Trump.

Some claim “shy Trump voters” concealed their views because they feared disapproval or even scorn. If most around you are persuaded one way, few openly refute them. They may be your friends, colleagues or bosses. The academic term is “social-desirability bias”: at least verbally, most adhere to majority norms.

How many Trump supporters would tell pollsters that they disliked seeing a woman — a Black woman, no less — become vice president? None, but Kamala Harris’s candidacy was a strong negative for many.

Why then does electoral polling sometimes work well? The answer, I propose, is that as social and political stress goes down, candid responses go up. Respondents not caught in a polarizing time little fear disapproval and so openly state their true voting intentions.

All we need then are calm times, something no law or constitutional amendment can provide. Researchers measure stress by asking a series of questions or panel studies ranking political statements. Polls could include such measures with survey results.

The clearest indicator, which we have just been through, is the degree of personal animosity expressed by at least one candidate. Hate-filled, abusive language both demonstrates and deepens society-wide tension. Panel studies can identify and rank abusive statements.

When was the last time we had a relatively stressless atmosphere for voting? Perhaps the 1976 Ford-Carter contest, both polite men who did not revile one another. Eisenhower and Stevenson in 1952 and 1956 were models of restraint and decorum. Polling predicted these outcomes, but we may never enjoy such tranquil times. Too much economic, social and demographic change.

Another way to measure political stress is by turnout. Post World War II, U.S. election turnouts declined to sometimes under 50%. People just didn’t care enough to vote. Some interpret this as a sign of basic contentment. The massive turnout upsurge this year — by both parties — shows angry discontent, a political fever afflicting many.

At a minimum, pollsters should announce: “In times like these, responses are apt to be off by several percentage points, so take these findings as approximate.” But they could also suggest whose poll numbers are overstated and establish a correction such as ones already used to adjust oversampled urban and undersampled rural respondents.

We should use surveys from late in the COVID cycle so that all know about it. We want to measure how many disbelieve it and correlate that with the Trump vote. Which comes first, virus-denying or pro-Trump inclinations? Does changing one change the other?

Combining two variables — the level of stress, which has climbed since at least Obama’s election, and who is doing the most verbal attacking — might indicate who is being overrated and by roughly how much. The trouble here, of course, is how can you tell if voters are drawn to or repelled by the verbally abusive attacker? Maybe they turn to the calmer personality.

For this, I think, one must abandon quantified tools and turn to personal experience, face-to-face interviewing (which pollsters avoid because it’s too expensive) and focus groups where typical citizens interact with each other. Some denounce “anecdotal” evidence as hopelessly unscientific, but good anecdotes beat bad quantified data. For weeks, MSNBC’s Joe Scarborough discounted Biden’s high Florida poll numbers. The former Florida congressman knew better, and Florida went solidly for Trump.

Indirect indicators may also provide tips. One unscientific journalist predicted Truman’s victory based on who purchased which products. In 1948, Truman “gave ’em hell”; Dewey was more restrained. Truman won by an upset 4.5 percentage points. In 2020, Trump MAGA caps and T-shirts vastly outsold Biden-Harris articles. Trump lost, but not by as much as projected.

Calm times produce moderate opinions and elections, allowing polling to work. Tense times vitiate accurate polling amid nasty polarization and incivility. So, unless President Biden miraculously heals us, polls will still be off by several percentage points. Recall that the Greek city-state, the polis, invented and required politeness, without which the polis was doomed.