Racism is a complicated and often misunderstood concept. However, specific to the 2016 electorate an increasing body of literature (I have referenced a number of those elsewhere) suggest that racial resentment, anti-immigrant sentiments and sexism are major predictors for Trump voters. That does not mean that all of them harbour racist tendencies, but it is a stronger indicator than other parameters typical for protest voters (such as distrust for politicians, dissatisfication with democracy) or economic hardship. A study from Mutz showed, for example that folks actually losing jobs or were impacted by stagnant wages were not the ones most likely to vote Trump. Rather, folks that felt under threat by social change (e.g. those that think that white folks are more discriminated against). The studies also imply that actual economic improvement will not change sentiments and indicates that the "left-behind-theory" most commonly presented by the media will not really address the needs of a significant size of the voters. Other studies also found that anti-establishment sentiments ("drain the swamp", hostility against "elites")
While it does not mean that it is a third (or more or less) of Trump voters, but it means that they are biggest coherent group driving the election for Republicans. There is of course a big chunk of partisanship voters, who will vote Republican regardless of the candidate, those may or may not fall into the above group, but certainly they are not sufficiently adverse to them, nor are they a viable group for recruitment for the opposing party, either.
Thus, these sentiments are major drivers of elections , even if it may be considered distasteful to be discussed publicly, compared to other narratives. This poses a fundamental problem in politics as it there are likely only a limited number of ways to engage with these voters, which, as the stats show, were key to the 2016 election.
One, is to pretend that these issues do not exist or engage in alternative narratives, such as the "left-behind" or "we are better" rhetoric (which to some degree seems to be the case with the Democratic party) and pretty much lose this voter potential. The basic idea is probably to split off those that vote Republicans for other reasons, but it is unclear whether that would be numerically relevant (or even possible).
Another one is to engage and demonstrate overlapping values (which happened in 2016 and it appears to be even stronger in 2020 in the Republican party) and thereby capture that voter base. That is not really that new, it is more that at least overt anti-immigrant sentiments have been retreating from the Democratic party (which makes them unvotable for said part of the electorate). That has the advantage of not rallying the base, but also recruit otherwise Democratic voters which harbour similar sentiments (a bit of that was observed also under Obama).
There are indeed papers looking into vote switching and found that in that group again, racial and immigration attitudes were the key factors (Reny et al. Pub Op Quart, 2019 83:1) and that these attitudes sort White voters into increasingly polarized groups.
However, Democrats cannot (luckily?) play the same game, as they would then more likely lose a significant chunk of minority voters, which have become a crucial voter base for them.
In other words, playing up racial sentiments are a win-win game for Republicans and lose-lose for Democrats (even mentioning racism in whatever context is often enough to shore up resentment). This, incidentally, is a sentiment that Bannon picked up and tries to sell hard as election strategy (not only in the US, but also in Europe, where we see similar mechanisms at play).