In this work we explore the possibility of using COSMIC/FORMOSAT-3 radio occultation profiles (ROP) to establish a statistical model of the deviations that can be expected between the monthly median values of N m F2 and h m F2 computed with the International Reference Ionosphere (IRI) and the actual values of these parameters. The actual values are retrieved from the ROP after an interactively re-weighted Least Square fit that, complemented with a statistical test, allows filtering of unreliable data and estimating the errors of the retrieved values. The differences between the retrieved values and the monthly median values computed from IRI are interpreted as the superposition of a systematic bias (attributed to both, IRI and ROP), random errors in ROP, and the day-to-day variability, which is unaccounted for by IRI. This variability is described with a five-dimensional function that depends on: the month, the solar activity, the geomagnetic conditions, the modip latitude, and the local time. Empirical values of this function are estimated in the form of regular grids.Since this research is restricted to low solar activity and quiet geomagnetic conditions, the grid is reduced from five to three dimensions: month, local time, and modip (modified dip latitude). We found that the standard deviation of the day-to-day variability varies according to (in percent of the monthly median value computed with IRI): (i) N m F2 at noontime: ±10% to ±30% with maxima over the northern and southern peaks of the Equatorial Anomaly; (ii) N m F2 at midnight: ±20% to ±45%, with the greatest values in the equatorial region during the months of May and September; (iii) h m F2 at noontime: ±2% to ±10% with minima over the modip equator; and (iv) h m F2 at midnight: ±3% to ±11% with the greatest values in the equatorial region from January to May and from September to January.