The paper studies a random variable built from the binomial distribution, which describes the number of successes in a series of coin flips. Specifically, if X counts those successes, the authors look at Y, the absolute deviation of X from its expected value. They want to understand the "absolute moments" of Y, which are averages of Y raised to integer powers, and which show up naturally in probability theory and statistics as measures of spread or error.
The main technical contribution is a clean, exact formula for every odd-powered absolute moment of Y. Rather than requiring an infinite series or numerical approximation, the formula involves only a finite combination of two types of quantities: the probability that X lands exactly at or near its mean (called "central masses"), and the probability that X falls in the tail beyond that point ("central tail probabilities"). The authors show that the coefficients in this formula obey a neat bookkeeping rule and always contain the factor (q minus p), where p and q are the two complementary probabilities of success and failure. This means that when the coin is fair, many terms vanish automatically and the formula simplifies considerably.
As a payoff, the authors use their formulas to derive precise approximations for the median of two well-known continuous probability distributions, the beta distribution and the gamma distribution. The median of the beta distribution does not have a simple closed form, so having a systematic expansion in powers of the sample-size parameter is genuinely useful. They recover and extend a known result due to Choi for the gamma median as a special limiting case. The work thus connects combinatorial identities about the binomial distribution to practical approximation formulas for continuous distributions used widely in statistics and Bayesian inference.