The skin effect does not depend on any mysterious underlying forces. It may be completely predicted based on one simple, lumpedelement equivalent model called the concentricring model.
Table 2.5. Assumptions Made by z BETTER
Conductor Geometry 
L i (0) 

Round wire of any radius not in close proximity to a return path ( k p = 1) 

Wire of any convex cross section with perimeter p and area a 

Rectangular wire of size w x t 

Square wire of size w x w 

Very thin rectangular wire, t << w 
Imagine a round conductor divided lengthwise into concentric tubes, like the growth rings on a tree trunk. In this model the current proceeds absolutely parallel to the wire's central axis. Because current naturally passes straight down the trunk, parallel to the dividing boundaries between the rings, you may insulate the rings from each other without affecting the circuit. You now have a collection of n distinct, isolated conductors, shorted together only at each end of the trunk.
You may now separately consider the inductance of each ring. The inner rings, like long, skinny pipes, have more inductance than the outer rings, which are fatter. You know that at high frequencies, current follows the path of least inductance. Therefore, at high frequencies you should expect more current in the outer tree rings than in the inner. That is exactly what happens. At high frequencies the current crowds into the outermost rings. At low frequencies current partitions itself according to the resistance of each ring, while at high frequencies it partitions itself according of the inductance of each ring.
This simple concentricring idea motivates the idea of a redistribution of current at higher frequencies. What it doesn't do, however, is properly indicate the magnitude of the effect. The skineffect mechanism is far more powerful than just the ratio of individual concentricring inductances. Mutual inductance between the rings actually bunches the current much more tightly onto the outer rings than you might at first imagine. The general setup for constructing a treering model, including the mutual inductance, is discussed in the article "Modeling Skin Effect." The concentricring circuit model, if taken to an extreme (hundreds of thousands of rings), properly predicts both skineffect resistance and skineffect inductance, at low and high frequencies, for a circular conductor.
2.9.1 Modeling Skin Effect
2.9.2 Regarding Modeling Skin Effect
Email correspondence received July 17, 2001 SE HO YOU writesYou have shown an inductance matrix and indicated that since all the entries in the right column of the matrix are the same, ring number 2 concentrates all its flux into the space between ring number 2 and the shield. Would you please explain why flux concentrates around ring number 2? The behavior of the flux comes first, I think. Then, the inductance matrix should be just a mathematical expression of how nature works. Could you explain more physically? 
ReplyEddy currents flowing in the outer ring create a magnetic shield through which flux cannot penetrate . The shielding effect of the outermost ring therefore prevents flux from reaching the inner rings. Receiving no electromagnetic impulsion from changing flux, no current flows on the inner rings. At low frequencies the eddy currents in the outer ring are impeded by the resistance of the copper , so the shielding effect is imperfect. With an imperfect shield, some magnetic fields do reach the interior and some current does indeed flow on the inner rings. As you go to higher and higher frequencies, however, the shielding effect becomes more pronounced. The improved shielding effect successively robs the inner rings of more and more flux (and current). 
Fundamentals
Transmission Line Parameters
Performance Regions
FrequencyDomain Modeling
Pcb (printedcircuit board) Traces
Differential Signaling
Generic BuildingCabling Standards
100Ohm Balanced TwistedPair Cabling
150Ohm STPA Cabling
Coaxial Cabling
FiberOptic Cabling
Clock Distribution
TimeDomain Simulation Tools and Methods
Points to Remember
Appendix A. Building a Signal Integrity Department
Appendix B. Calculation of Loss Slope
Appendix C. TwoPort Analysis
Appendix D. Accuracy of Pi Model
Appendix E. erf( )
Notes