The pundits and pollsters have been criticized ad nauseam for failing to correctly predict the 2016 presidential election. Granted, they didn’t get everything wrong. The popular vote tallies matched the predictive models more closely than the Electoral College projections, which saw Hillary Clinton winning a decisive victory. Nonetheless, it is has been a time of many apologies. Both political science as an academic discipline and professional pollsters have a bullseye on their back for their failures.

Social scientists should recognize and acknowledge their mistakes. But we should first be clear about what these mistakes were. More pressing than flawed models and a failure to predict outcomes is the political science discipline’s complicity in limiting our political imagination. Mainstream political science views its mission in a highly depoliticized, technocratic manner that results in a blinkered understanding of the range of political possibilities. Taking on that vision is more important than ever in the age of Trump.

Polling has become gospel in American politics. Without polling and reliance on datasets, it is nearly impossible to make a respected political argument. This is unfortunate, because polling by its nature is flawed. It does not provide a neutral snapshot of voters’ understandings of politics. Instead, polls reflect the dominant narratives at a moment in a way that naturalizes key controversies.

In contrast to the assumptions of most poll- and survey-based studies, political psychologists demonstrate that exposure to campaigns (both directly and through the media) influences voters in sophisticated ways, and does not merely “reinforce voters’ preexisting partisan loyalties.” People don’t exist in a vacuum; polling itself shapes the narratives they adopt to understand and make decisions about policy.

This past year polling numbers were used to reinforce conventional political strategies precisely along these lines. For over a year, experts assured us that Sanders was unelectable and urged the nomination of the “safe” candidate. While acknowledging that Democrats had gone “more liberal” in recent years, pollsters argued that voters were not yet liberal enough to elect Sanders. Such polling-driven arguments foreclosed the real possibility that surging support for Sanders might not require voter self-identification as “more liberal.”

Polling data couldn’t adequately capture a comprehensive picture of voters’ thoughts and orientations, which are after all nothing more than preferences at a single moment and depend almost entirely on how various questions are framed.

Polls (and media reliance on polls) also naturalize and delineate the potential range of political outcomes. People are seen as consumers, with preexisting, ordered preferences that drive their political behavior; they are presented with a limited range of choices and issues and must respond in a circumscribed, preordained way. Pollsters, experts, and political scientists then use this data as a definitive representation of the electorate rather than for what it is — imperfect timebound data points that require triangulation with other forms of information.

Political scientists and pollsters, relying on their existing models and methods, could only dismiss the Sanders campaign as a long-shot distraction. Sanders lacked the name recognition, party connections, resources, and other standard markers of a winning candidate. The fact that Bernie Sanders, not Hillary Clinton, filled stadiums and had supporters waiting for hours in cold weather to hear him speak may not have been quantifiable — but in retrospect it was more meaningful than anything represented by polling at the time.

Even in the election postmortem, there is strong pushback against claims that Sanders, or a more socialist, left-populist candidate in general, could have beat Trump. To squelch this idea pundits point to exit polls that demonstrate that the number one issue reported by Trump supporters was immigration, followed by terrorism. Sanders did not run a xenophobic campaign, so by extension we’re told his message would not have gained traction among Trump supporters.

As socialists, we must reject this limited poll- and data-driven thinking. Just because many Trump supporters checked off the box for immigration on exit poll questionnaires does not that mean that they could not join a broad, multiracial, working-class coalition to fight for progressive change in the future. Most Trump supporters also felt the economy was doing poorly, in contrast to the Democratic narrative praising Obama’s recovery. This presents a clear opportunity to introduce a progressive, redistributionist narrative that moves beyond poll-centered politics.