Studies of how different work practices affect organizational performance h
ave suffered from methodological problems. Especially intractable has been
the difficulty of establishing whether observed links are causal or merely
reflect pre-existing differences among firms. This analysis uses a national
probability sample of establishments, measures of work practices and perfo
rmance that are comparable across organizations, and, most important, a uni
que longitudinal design incorporating data from a period prior to the adven
t of high-performance work practices. The conclusion most strongly supporte
d by the evidence is that work practices transferring power to employees, o
ften described as "high-performance" practices, raise labor costs per emplo
yee, suggesting that they may raise employee compensation. Higher compensat
ion is a cost to employers, although some statistically weak evidence point
s to these practices raising productivity. The authors find little effect o
f high-performance work practices on overall labor efficiency, which they m
easure as the output per dollar spent on labor.