There’s more heat than light in this drill-down of the NBC/Wall Street Journal poll out this morning, but the heat’s worth checking out anyway. Allahpundit noted the national result yesterday, which had Joe Biden up ten points in the overall sample of 1,000 registered voters conducted between Thursday and Saturday.

National polling is relatively meaningless, however. What about the battlegrounds? NBC/WSJ surveyed 800 more registered voters in twelve states, and found a closer race:

In the 12-most competitive battlegrounds — Arizona, Florida, Georgia, Iowa, Maine, Michigan, Minnesota, Nevada, New Hampshire, North Carolina, Pennsylvania and Wisconsin — Biden is ahead of Trump by a combined 5 points, 51 percent to 46 percent, according to a survey of 800 additional registered voters in those states interviewed Sunday by NBC/WSJ pollsters.

Biden’s lead in these states was 6 points in our Oct. 29-31 NBC News/WSJ poll, 51 percent to 45 percent, although that movement is within the margin of error.

Important context, however: Trump won these same states by a combined 2 points in 2016, 49 percent to 47 percent.

There are reasons to take this poll with a grain of salt, but let’s stick with the numbers first. According to the data, NBC/WSJ asked the same question of battleground voters three weeks ago (10/9-12) and showed Biden up by ten, 52/42, which is well outside the margin of error. If we can rely on this poll, at least for trending, it appears Trump has begun to surge in October — although that might not have been early enough, given the mail-in balloting and early in-person voting that began in September.

Another data point isn’t so cheery. The pollster asked what chances each candidate had of getting the respondents’ votes. For Biden, 43% said “no chance at all,” while 49% said the same thing about Trump. That would suggest a very narrow path to victory for Trump at best even in these battleground states he already won.

That’s only a problem if one trusts this poll. Perhaps it’s accurate as far as it goes, but the sample size is a huge problem. Eight hundred registered voters might be an acceptable size for any one of these states, but not for twelve states lumped together. It comes to just under 67 voters per state interviewed by NBC/WSJ, which is an absurd number for any predictive value. Plus, let’s note that this is not a likely-voter model, which would usually be more predictive, if pollsters could figure out who’s likely in a cycle such as this.

This is just another straw in the wind, but even if it’s accurate, it might not be indicative. The actual voters in this election might be different than the registered-voter model; in fact, given the disparity of the ground game, you can pretty much bet that it will be. That, plus the inherently higher failure rates of mail-in ballots, means that the end result will likely be much closer than this.