Why all Armenian polls failed spectacularly: garbage in, garbage out
You are trying to make sense out of very noisy data, and whatever you do with this garbage data will not improve it. All of publicly available polls are problematic.
Armenia’s 2026 election exposed a crisis in political polling that extends far beyond statistical error. Every major polling organization-EVN Report, IRI, and MPG-dramatically underestimated Civil Contract, with gaps ranging from 16 to 25 percentage points. Even the best-performing poll (MPG, with a forecast accuracy score of 14.3) was wildly off. Hrant Mikaelian traces this failure to three fundamental problems that plague Armenian polling.
First, publicly available polls are inherently corrupted by the interests of whoever funds and disseminates them. Unless one believes polls are commissioned by philanthropists motivated by pure civic virtue, the reasonable assumption is that polls are designed to affect public opinion, not merely report it. Who paid for the poll, and what outcome did they prefer? This question shapes which organization gets hired, when the poll is conducted, and how results are presented to media outlets.
Second, timing matters enormously. Pashinyan’s support surged dramatically after the European Political Community summit in early May and the military parade one week before the election. A poll from March captured a very different moment than one from late May. The EVN Report poll from March 31, for example, showed Civil Contract at 33.6%, but the actual result was 49.8%. Much of that gap reflected genuine shift in campaign momentum, not polling error alone.
Third, and most critically, the analysis and data processing distort the raw numbers. Multiple pollsters disclosed high refusal rates (75-85%), yet handled this distortion differently. The interpretation of raw data-the weighting, the treatment of non-response, the demographic adjustments-determines the final result far more than the raw figures themselves. Some polls attempted to correct for fear-based non-response; others did not. Unpublished polls, which included richer metadata and more sophisticated processing, came closer to actual results, but remained inaccessible to the public. The paradox: the most methodologically rigorous polls never see daylight, while flawed public polls drive media and opposition strategy.