The Poll Herding Problem: When Independence Is an Illusion

The Question

Seven major polling organizations released presidential approval numbers within a single week. All seven fell between 37.4% and 39.2%, a range of 1.8 percentage points. Given the different methodologies, sampling frames, likely voter screens, and weighting assumptions employed by these organizations, is this clustering evidence of genuine convergence in public opinion, or is it evidence of something else entirely?

WHAT HAPPENED

• Gallup: 38.2% approve
• Reuters/Ipsos: 37.8% approve
• Morning Consult: 38.6% approve
• YouGov/Economist: 38.1% approve
• Civiqs: 38.4% approve
• Monmouth: 37.4% approve
• Quinnipiac: 39.2% approve

THE PICTURE IN OUR HEADS

The public sees seven polls telling the same story, which creates an impression of certainty. If every poll agrees, the result must be reliable. This is intuitive and wrong.

Independent polls using different methods should produce different numbers. When they produce the same numbers, either public opinion has genuinely converged to a single point (which is statistically implausible), or the polls are not truly independent.

THE MACHINERY

Poll herding occurs through several mechanisms, none of which require conscious deception. A pollster whose result differs significantly from the consensus faces professional risk. An outlier poll generates scrutiny of methodology. The safest result is the one that matches the other results.

The mechanisms are subtle: adjusting likely voter screens to produce a more "reasonable" topline, weighting demographic cells to match assumptions rather than raw data, or simply choosing not to release polls that look anomalous. Each decision is defensible individually. Collectively, they produce artificial consensus.

The historical average weekly spread for presidential approval polls is 4.7 percentage points. This week's spread of 1.8 points is less than 40% of that average. The standard deviation of this cluster (0.62 points) is less than one-third the historical norm. This degree of convergence is more consistent with herding than with genuine agreement.

THE COMPETING FRAMES

Two interpretations are available. The "convergence" frame treats the clustering as evidence that polls are getting better, their methods are converging on truth. The "herding" frame treats the clustering as evidence that polls are getting more alike, which is a different thing entirely.

If the herding thesis is correct, the true uncertainty in the president's approval rating is substantially larger than the polling cluster suggests. The real range might be 35–42%, not 37–39%.

WHAT THE PUBLIC ACTUALLY SEES

The public sees consensus. Consensus creates confidence. Confidence reduces scrutiny. The irony is precise: the closer the polls agree, the less independently informative each one is, and therefore the less reliable the consensus should be treated.

POLLERBULL SIGNAL

• What moves odds: If one pollster breaks from the herd and publishes a number outside the 36–40% range, watch whether the outlier or the consensus proves more accurate over subsequent weeks.
• What would falsify this: If all seven polls are using genuinely independent methodologies and the convergence reflects real opinion stability, the approval rating will remain in this range for at least four more weeks regardless of external events.

SOURCES

• Individual pollster releases, February–March 2026
• FiveThirtyEight, historical presidential approval poll spreads
• American Association for Public Opinion Research, methodological standards