One of the questions at the heart of the Democratic presidential contest is whether the race has shifted in favor of Sen. Bernie Sanders over the past few months. There's no question that Sanders has improved against Hillary Clinton over the course of the past 12 months, narrowing her once-massive lead to single digits.

Even since the beginning of the year, Sanders consistently cut into Clinton's lead in the RealClearPolitics average of polls. When voting began on Feb. 1, he trailed her by 14.4 points. By the end of the month, he was within 5 -- about where he is now after closing the gap and then watching it grow.*

The trend at the very end of that chart, though, contradicts one of Sanders's arguments to superdelegates, that he is the candidate who Democrats want as their nominee now, versus who they wanted three months ago. The idea is that his consistent state wins are a marker of a shift in the electorate that should be rewarded at the convention in Philadelphia. This isn't Sanders's only argument to superdelegates, but it's one that his campaign and his supporters have made.

That's surprisingly hard to evaluate. It's hard to compare a series of contests in a number of different states and taking place under a slew of different conditions and to evaluate the extent to which the performance of one of two candidates has changed. This morning we noted the ways in which the contests in the generally similar states of Kentucky and West Virginia differed, giving Sanders a blow-out win in the latter and allowing Clinton to run even in the former. There was a week's difference between the two, but two very different contests.

Sanders has regularly equated these state wins, like the patch in late March where he won a string of contests, as evidence that the race was shifting. That's not the case.

At the end of February, FiveThirtyEight outlined what demographics suggested about who would win the rest of the Democratic contests if the two were tied nationally, and by how much. Because Clinton maintained a lead in national polling, she overperformed against the predictions, but the broad strokes were accurate: States favoring Sanders on demographics were states where he did better. Sanders's "momentum" in those wins at the end of March was a function of those states being ones where demographics gave him mostly double-digit leads.

The FiveThirtyEight predictions had Sanders winning West Virginia by 17, Oregon by 20 and Kentucky by 3. He won West Virginia by 16, Oregon by 12 and tied in Kentucky -- meaning Clinton outperformed slightly, which you'd expect since she leads nationally. In the first three contests, they predicted a 19-point Sanders win in Iowa, a 32-point victory in New Hampshire and a tie in Nevada. Clinton outperformed by more in those first two contests -- but, then, her national lead was bigger then, too.

Meanwhile, demographic splits have been remarkably consistent. We looked at this a month ago; polling from The Post and ABC News shows that while the overall race has closed, demographic groups that preferred Clinton nine months ago prefer her now, and vice versa. Two examples are age and gender.

This carries over into exit polling from contests where we have data. Exit polling groups are generally fairly broad, and it can be hard to compare between disparate contests. If we look at how Sanders performed with each demographic relative to his overall performance in the state, we see that those numbers haven't changed much over the course of the campaign. The last two columns on each of these graphs (which you can change) are the average of the first four and last five contests for which we have exit poll data.

Sanders's performance with groups relative to his state margins hasn't really changed very much from the start of the campaign to now, especially relative to the margin of error. His average in the last five contests versus the first four has improved by 3 points or more with moderates, whites without a college degree and people ages 30 to 44 or 65 and up. It has worsened by that much with non-white voters without degrees.

There have been instances in which Sanders has outperformed polling. In Oregon, for example, a poll showed Clinton up 15 points last week -- but that was one poll that, even at the time, seemed like an outlier. He beat the average in West Virginia -- an average based on two polls. In Indiana, he beat the average by 11 points. In New York, Clinton beat it by 4; in Maryland, by 6.

The best data we have for what we'd have expected the races to look like, then, may be the combination of the national margin with demographic predictions. That undermines the argument that Sanders is doing better, since things appear to be lining up with the way we'd expect. If demographics are the best predictor and demographic splits haven't changed much, not much has changed.

That this is hard to evaluate plays to Sanders's advantage. His winning states feels like it's based on momentum, even if there's little evidence that this is the case. But there's another flaw to that argument. Of the last five contests, Sanders has won three (losing Kentucky and Guam). Of the last 10, he's only won five.

National polling suggests that Sanders continued to improve against Clinton for most of the year. The state-by-state vote, though, is far too murky to say with certainty that much has changed. Instead, the evidence at hand suggests that any improvement since the end of February has been relatively minor.

* Reuters' polling has been more consistent since Feb. 1.