It's the chocolate that'll getcha.

Interesting question, it reflects on the discussion about news media bias, and "versions of the truth."

(1) 200 Million from Politico quoting Target Smart a vendor to the Democratic Party from 2016 here: http://www.politico.com/story/2016/10...
"There is no current national database of voter registration because each state independently runs its own election."

[Well, actually, there apparently is, now. It is not surprising. Back in the early days of computing, in East Lansing, I met Mark Grebner (Wikipedia here: https://en.wikipedia.org/wiki/Mark_Gr... ) who specialized in voter databases.]

"But TargetSmart, a Democratic political data firm, told POLITICO that the country passed the 200 million threshold in recent days as North Carolina, New Hampshire, Nevada and New York reported new voter numbers.

Tom Bonier, CEO of TargetSmart, said national registration now stands at 200,081,377 voters."

(2) US Census Bureau November 2016 many tables here: https://www.census.gov/data/tables/ti...

157,596,000 self-reported registered voters

(3) From New York Magazine, Oct 19, 2016 http://nymag.com/daily/intelligencer/...

"Sadly for Donald Trump, all of these new voters are not white men. In fact, it appears that the new voters heavily lean Democratic, with 42.6 percent registered as Democrats and only 29 percent registered as Republicans — just 0.6 percent more than those who identify as independent. The major growth engine of the trend has been the United States’ booming Hispanic populations, which are expected to overwhelmingly favor Hillary Clinton.
[...]
"Still, these numbers are not definitive. They all come from TargetSmart, a Democratic political-data firm. Because there is no central national record of registered voters and all voter records are handled at the state level, it is left up to independent firms like TargetSmart to tabulate national voting trends."

Note the key word "potential."

"In the early evening of June 12th, UpGuard Cyber Risk Analyst Chris Vickery discovered an open cloud repository while searching for misconfigured data sources on behalf of the Cyber Risk Team, a research unit of UpGuard devoted to finding, securing, and raising public awareness of such exposures. " -- https://www.upguard.com/breaches/the-...

"The RNC data repository would ultimately acquire roughly 9.5 billion data points regarding three out of every five Americans, scoring 198 million potential US voters on their likely political preferences using advanced algorithmic modeling across forty-eight different categories."

I mentioned above East Lansing political consultant and politician, Mark Grebner. When I ran for Congress in 1992, I went to him for voter lists. From Wikipedia cited in the intro: "In an article about local and national campaign consultants in the early years of electronic data science, The Wall Street Journal described the work necessary to transform raw voter data into a mailing label targeted to specific voters: "Mr. Grebner painstakingly gathers lists of local voters from hundreds of Michigan township and municipal clerks, many of whom don't have computers and keep only hard-copy records. He enters the names in his battery of desktop computers, then matches them to addresses and other data." [Sterba, James (September 1, 1992). "Democracy Inc.: Politicians at All Levels Seek Expert Advice, Fueling an Industry" (Eastern Edition). The Wall Street Journal]

Just for example, you voted in the presidential election of 2012, but not in 2016. You never voted in a local election. Depending on your state, you likely dropped off the voter rolls. But to people who construct the databases, you are still a potential voter. Say you voted in a midwest township in 2012, but then moved to New England and voted in 2016. You might be counted twice by the pollsters until they combed through the databases to normalize them. With no other way to identify you - no social security number, for example: not used as voter ID - you could be counted twice easily and remain so tallied.

The salient fact is not whether the number is 158 million or 200 million but that the records were unsecured.