I’m not sure exactly what you mean by “the median class when ranked falls at zero”. But suppose that the median class is from 0 to 2, say, so that its midpoint is 1, and that its frequency is 16 (out of 30 in the dataset). Then the class boundaries are -1/2 to 2 1/2, so that L = -1/2, N = 30, F = 0, f = 16, and C = 3. The formula gives m = L + [ (N/2 – F) / f ] * C = -1/2 + [ (30/2 – 0) / 16 ] * 3 = 2.3125 (that is, 2 5/16). This is, of course, only an estimate of the true median, based on the assumption that these 16 people have values evenly distributed from -1/2 through 2 1/2. Since the values are actually 0, 1, and 2, the actual median could in principle be 0, 1, or 2, depending on the distribution.

Note that if the first class is the median class, then f has to be at least N/2 so that this one class will contain at least half the data. You would definitely prefer to use the raw data and find out how many actually are zero, because the classes are far too wide. If more than half of your people attended no training sessions, then the median is indeed zero.

]]>