A bathtub curve is a plot of a hazard function that starts high and decreases, remains low for a period, and then increases. In reliability analysis these regions model early failures, a period of relatively stable failure rates, and wear-out, respectively. The high ends and low middle resemble the side profile of a bathtub, which gives the curve its name (NIST/SEMATECH). Klutke et al. (2003) discuss limitations of the model, especially the assumption of a declining failure rate near time zero.
For example, the figure shows the hazard function
where
is time. The exponential term dominates at early times, the constant term provides
a baseline, and the quadratic term dominates at late times. The survival
function associated with this model is
For a Weibull distribution with shape and scale
, the hazard
function is proportional to
for
. It is decreasing for
, constant for
, and increasing for
. Thus a single Weibull
distribution cannot produce a bathtub curve. Glaser (1980) gives criteria for
characterizing bathtub-shaped and other hazard functions.
Hjorth (1980) gives a family of lifetime distributions whose hazard
functions can be increasing, decreasing, constant, or bathtub-shaped.