Given a reference group k+1, the logit of j level and the probability could be computed
with equation (6) and (7) respectively.
pj
(6) log----= Bj x for j = 1,∙∙∙, k
pk+1
(7)
exp(B j x)
p, = -=,----------
∑exp( Bjx)
j
for j =1,∙∙∙,k+1
When the cumulative logit function is implemented, the reference category is not
fixed and changes with the level of the event. The logit and the corresponding probability
would be computed with equation (8) and (9). Parameters in logit models are interpreted
(8) log-JP-, log pι + p2 L log Pι + P2 +^^^ + p^
1 - p 1 1 - (p 1 + p 2 ) 1 - (p 1 + p 2 + ∙∙∙ + pk )
(9) pj = -XpBX--pj -1 forj = 1,∙∙∙, k
∑Bjx
exp j
j=1
as corresponding increases in logit of the response variable to a unit increase in
explanatory variables. If explanatory variables are categorical, such logit change reflects
the relative impact of a level to the reference level. As indicated in equations (7) and (9),
the probability of specific behavior choices could be derived from estimated parameters.
Model Implementation
The demand analysis in this study consists of: the existing demand, the demand potential
from per capita consumption increase, the demand potential from new consumers, and the
demand changes related to season and occasions. In each case, we start with an initial set