Why not calculate turnout rates as percentage of registered voters?
If the goal is to calculate turnout rates for eligible voters, isn't the number of people registered to vote the eligible population? This may seem reasonable, but voter registration statistics are not exactly what they may seem.
Voter registration rolls contain "deadwood" -- people who are registered at an address but no longer reside there for whatever reason. To alleviate the burden of registering each election -- a tactic that was used in the past to suppress voting among certain groups -- federal law requires that a registered voter remain on the voter registration rolls for no less than two federal general elections since the last time the person voted. Persons can only otherwise be removed from the voter rolls through a process called "purging" if they notify the post office of a change of address or fail to respond to a written notice from election officials (there are other ways a registrant may be purged -- these are two common methods). Purging practices vary among localities and vary over time, and thus the amount of deadwood may vary, rendering voter registration turnout rates non-comparable across time and space.
The way by which eligible persons can register to vote varies across states. North Dakota does not have voter registration. Other states have a policy known as "Election Day Registration" or EDR which allows individuals to register and vote on Election Day. In these states, voter registration as a turnout rate denominator is misleading since any eligible voter can cast a ballot if they so desire.
If one wants to know the effect of a policy like EDR on turnout, a consistent denominator like VEP is preferred to measure effects which would otherwise be confounded by purging practices.
Although voter registration turnout rates may be uninformative for comparison purposes, election officials often calculate turnout rates for registered voters. Some don't, for example, Minnesota reports turnout rates computing using the VEP statistics found here. Why do they do this?
From a practical standpoint, voter registration is a voter turnout denominator that election officials have readily at hand. In contrast, calculating VAP and VEP requires familiarity with data sources that election officials do not often work with.
Some states or localities have policies regarding the creation of voting precincts or allocation of election machines that depend, in part, on voter registration and turnout.