Earlier this week, the Washington Post partnered with Survey Monkey in an ambitious, if potentially fatally flawed, attempt to conduct a 50-state poll. Morning Consult released the third iteration of its own version this morning, with results that match up more closely with other polling at the national and state level. It provides fewer surprises, but does indicate that Hillary Clinton hasn’t clinched the race over Donald Trump — not yet, anyway:

Hillary Clinton would top Donald Trump 321-195 in electoral votes to clinch the White House if the election were held today, according to an extensive Morning Consult analysis of registered voters.

But the 2016 presidential race is far from over. The leading candidate is within the margin of error in nine states: Arizona, Florida, Georgia, Iowa, Maine, New Mexico, Nevada, North Carolina and Pennsylvania. When those states are removed from the count, the former secretary of State garners 258 electoral votes to the New York businessman’s 164.

The analysis is the third time Morning Consult has provided a complete look at the electoral college map. Voter sentiment has shifted in a number of key states since our last survey, which was released just before the political conventions. Our analysis also provides a snapshot into how Libertarian candidate Gary Johnson and the Green Party’s Jill Stein would affect the Electoral College count for Clinton and Trump.

So without the leaners, Hillary comes within 12 electoral votes of victory. New Mexico seems highly unlikely to flip to the GOP, but it’s hardly the most significant of these states. The current RCP average for Pennsylvania has Hillary up 6.5 points, a much wider gap than MC’s virtual tie with Hillary up 0.9%. That’s 20 electoral votes right there, which means Trump has to win Pennsylvania in this model.

Since I criticized the WaPo/SM poll’s methodology, let’s take a look at the MC model too.  They polled 18,000 registered voters over the entire month of August, a similar time frame as the WaPo/SM survey with a much smaller sample. This has the same potential problem identified in the WaPo/SM poll, which is this: what time slice does this represent? It’s relatively safe to take a month-long poll on a static issue, such as abortion, gun rights, immigration, and so on, because those issues tend not to have dynamic swings. An election, on the other hand, has a number of dynamic shifts — and August’s polling environment demonstrates that for the presidential race. The longer the survey period goes, the less likely that the results on one end will correspond with the results on the other end.

In other words, did Hillary have a 321-195 lead in the first week of August, or the last week of August? Does this poll reflect the reality of the electorate now or a month ago? Actually, if polling finished at the end of August, these results are a week old now, and don’t take into consideration events that might be impacting voter decisions now.

Apart from those concerns, the level of people having no opinion in all 50 states seems a bit high. In the four-way race, that ranges from 12-15%; in the two-way race, the range runs from 16%-21%. That introduces a significant amount of ambiguity to these results that suggest less reliability. The most recent YouGov poll does come close to those ranges, with 10% not picking a candidate in the four-way question and 14% in the binary question. But in YouGov’s Pennsylvania poll, only 9% didn’t pick a candidate in the 4-way question, while MC’s data shows 13.2% undecided, a difference outside the MC margin of error. Given Pennsylvania’s critical standing in the race, as well as the deviation between the MC results and the RCP average, some caution should be taken.

The results seem a little more solid than the WaPo/SM survey, especially regarding Texas. But we’re probably better off relying on individual state polling conducted within tighter time frames if we want to know the current status of the race.