The 2017 NFL draft begins tonight, and teams have been working hard to accurately estimate the value of draft prospects. The consequences can be especially dramatic for those teams considering a quarterback. In 1998, for example, the question was: Peyton Manning or Ryan Leaf? It’s hard to believe it now, but that was a tough decision.

This year’s decision involves Deshaun Watson and Mitch Trubisky, a comparison that evokes the complicated history of race and NFL quarterbacks. This includes a history of black quarterbacks being more likely to be asked to switch positions, being paid lower salaries than equally valuable white quarterbacks and being subjected to double standards.

Watson addressed this last year, arguing against being labeled a dual-threat, a quarterback that could run and throw. “That’s a code word. … People think, ‘Oh, he’s a black quarterback, he must be a dual-threat.’”

Our research suggests Watson’s concerns are well-founded.

We studied official NFL draft profiles and found substantial racial differences in the language used to describe quarterback prospects — differences that are consistent with established racial stereotypes. Perhaps most importantly, racialized language actually helps predict draft outcomes.

Racial stereotypes are common in draft reports about quarterback prospects

We analyzed the text of draft profiles hosted on the NFL’s website for all quarterbacks who participated in the NFL combine between 2008 and 2016. This constituted a total of 175 reports for 43 minority and 132 white prospects.

To identify racially differentiated language in this text, we used a statistical model of language that was originally developed to identify differences in how Republicans and Democrats discuss issues in Congress.

The results of this analysis are striking. Draft profiles discuss black and white quarterbacks using very different language. This language evokes the racial stereotypes noted in previous social science scholarship.

A white quarterback prospect is more likely to be discussed in terms of intangible internal qualities for which he himself is responsible. He is smart, displays intelligence, and understands the game. He is a leader with command of the huddle. He is consistent, calm, and poised. He is credited for his production. He is good or even outstanding. He appears to fit the prototype.

In contrast, a minority quarterback prospect is more likely to be discussed in terms of physical characteristics, to be judged erratic and unpredictable, and to have his successes and failures ascribed to outside forces. We learn about his hands, his weight, his frame, his body, parts of which are often either big or lean. He bolts prematurely, rather than stand in the pocket, or perhaps he hesitates before throwing dangerous passes. Less of a leader, he is asked to do things or given opportunities. His game has deficits. His footwork’s a mess. He has issues.

There’s no objective basis for these language differences

While physical words like big and weight are more prevalent in reports about minority prospects, one might object that perhaps minority quarterbacks just weigh more, or less, than white quarterbacks.

That’s not the case.

Because these prospects participated in the NFL combine, we have measures of things like their actual weight and height. But we found that a prospect’s actual weight did not predict whether his weight was mentioned.

But race is a different story. For example, we estimate that for a white prospect at the average weight of 223 pounds, there is a 6 percent chance that his report will use the word “weight.” For a minority prospect weighing 223 pounds, the chance is much higher: 27 percent.

Minority prospects, it would seem, are being evaluated by different criteria than white prospects are.

This matters for when players are drafted

Given the millions of dollars at stake, it could be that these are “just words” and don’t reflect any actual differences how teams evaluate quarterback prospects. But it appears that these words do matter.

The presence of racialized language in reports helps predict prospects’ draft position, even after accounting for how prospects performed in college performance and at the combine. For example, when words like issues are included in a minority prospect’s report, that prospect is predicted to be drafted about half a round later than otherwise.

In the unusual cases that reports include language that contradicts existing stereotypes, a prospect’s predicted draft position can drastically change. For example, when words related to intelligence or leadership are used in a minority prospect’s report, that prospect is predicted to be drafted roughly a full round earlier than otherwise.

So what about DeShaun Watson? His NFL draft profile describes him, yes, as a “dual-threat weapon,” as well as a “tremendous athlete” with “big, strong hands,” who was “baited into bad-decision interceptions.” On the other hand, it also describes him as a “tremendous leader and winner.”

It may come as further consolation that Trubisky’s profile starts by noting he, too, is a “dual-threat.”

Christopher Boylan is a PhD student in political science at Pennsylvania State University. Ryan McMahon is a PhD student in political science and social data analytics at Pennsylvania State University. Burt L. Monroe (@burtmonroe) is a professor of political science and head of the program in social data analytics at Pennsylvania State University.