Reimplementing Underscore’s _.reduce function by using the functionality of the _.each function

I’m working on problem for an upcoming bootcamp. Here’s where I’m at so far:

_.reduce = function (collection, iteratee, accumulator, context) {
if (arguments[2] !== undefined) {
    _.each(collection, function (element, key) {
      accumulator = iteratee.call(context, accumulator, element, key, collection);
    });
  } else {
    _.each(collection, function (element, key) {
      if (key === ???) {
        accumulator = collection[key];
      } else {
        accumulator = iteratee.call(context, accumulator, element, key, collection);
      }
    });
  }
  return accumulator;
};

I could write this longhand (iterating through objects || arrays manually), without using _.each. But a) that would spoil the fun and b) I get the feeling it’s possible as above. The issue is, in the instance that the accumulator argument isn’t provided as an argument, I need to iterate from the second value of the object || array, using the first value as the accumulator. But in writing the logic for this, I’m stuck on how to identify this first value, given it could be either an object property, or an array element. Hence the ??? in line 8. Hints welcome.

Cheers.